home *** CD-ROM | disk | FTP | other *** search
| MacBinary | 1999-03-08 | 34.9 KB | [TEXT/ttxt] |
open in: 
MacOS 8.1
|
Win98
|
DOS
browse contents |
view JSON data
|
view as text
This file was processed as: MacBinary
(archive/macBinary).
| Confidence | Program | Detection | Match Type | Support
|
|---|
| 10%
| dexvert
| MacBinary (archive/macBinary)
| fallback
| Supported |
| 1%
| dexvert
| Text File (text/txt)
| fallback
| Supported |
| 100%
| file
| MacBinary II, inited, Mon Mar 8 08:45:34 1999, modified Mon Mar 8 08:45:34 1999, creator SimpleText, type ASCII, 31229 bytes "KnowledgeMiner FAQ" , at 0x7a7d 4300 bytes resource
| default (weak)
| |
| 99%
| file
| data
| default
| |
| 74%
| TrID
| Macintosh plain text (MacBinary)
| default
| |
| 25%
| TrID
| MacBinary 2
| default (weak)
| |
| 100%
| siegfried
| fmt/1762 MacBinary (II)
| default
| |
| 100%
| lsar
| MacBinary
| default
|
|
| id metadata |
|---|
| key | value |
|---|
| macFileType | [TEXT] |
| macFileCreator | [ttxt] |
hex view+--------+-------------------------+-------------------------+--------+--------+
|00000000| 00 12 4b 6e 6f 77 6c 65 | 64 67 65 4d 69 6e 65 72 |..Knowle|dgeMiner|
|00000010| 20 46 41 51 00 00 00 00 | 00 00 00 00 00 00 00 00 | FAQ....|........|
|00000020| 00 00 00 00 00 00 00 00 | 00 00 00 00 00 00 00 00 |........|........|
|00000030| 00 00 00 00 00 00 00 00 | 00 00 00 00 00 00 00 00 |........|........|
|00000040| 00 54 45 58 54 74 74 78 | 74 01 00 00 00 00 00 00 |.TEXTttx|t.......|
|00000050| 00 00 00 00 00 79 fd 00 | 00 10 cc b3 09 84 fe b3 |.....y..|........|
|00000060| 09 84 fe 00 00 00 00 00 | 00 00 00 00 00 00 00 00 |........|........|
|00000070| 00 00 00 00 00 00 00 00 | 00 00 81 81 e6 fc 00 00 |........|........|
|00000080| 51 3a 20 57 68 61 74 20 | 64 6f 65 73 20 47 4d 44 |Q: What |does GMD|
|00000090| 48 20 73 74 61 6e 64 20 | 66 6f 72 3f 20 0d 41 3a |H stand |for? .A:|
|000000a0| 20 47 4d 44 48 20 2d 20 | 47 72 6f 75 70 20 4d 65 | GMDH - |Group Me|
|000000b0| 74 68 6f 64 20 6f 66 20 | 44 61 74 61 20 48 61 6e |thod of |Data Han|
|000000c0| 64 6c 69 6e 67 2e 20 49 | 74 20 69 73 20 61 20 73 |dling. I|t is a s|
|000000d0| 74 61 74 69 73 74 69 63 | 61 6c 20 6c 65 61 72 6e |tatistic|al learn|
|000000e0| 69 6e 67 20 6e 65 74 77 | 6f 72 6b 20 74 65 63 68 |ing netw|ork tech|
|000000f0| 6e 6f 6c 6f 67 79 20 75 | 73 69 6e 67 20 74 68 65 |nology u|sing the|
|00000100| 20 63 79 62 65 72 6e 65 | 74 69 63 61 6c 20 61 70 | cyberne|tical ap|
|00000110| 70 72 6f 61 63 68 20 6f | 66 20 73 65 6c 66 2d 6f |proach o|f self-o|
|00000120| 72 67 61 6e 69 7a 61 74 | 69 6f 6e 20 69 6e 63 6c |rganizat|ion incl|
|00000130| 75 64 69 6e 67 20 73 79 | 73 74 65 6d 73 2c 20 69 |uding sy|stems, i|
|00000140| 6e 66 6f 72 6d 61 74 69 | 6f 6e 20 61 6e 64 20 63 |nformati|on and c|
|00000150| 6f 6e 74 72 6f 6c 20 74 | 68 65 6f 72 79 20 61 6e |ontrol t|heory an|
|00000160| 64 20 63 6f 6d 70 75 74 | 65 72 20 73 63 69 65 6e |d comput|er scien|
|00000170| 63 65 2e 20 47 4d 44 48 | 20 69 73 20 6e 6f 74 20 |ce. GMDH| is not |
|00000180| 61 20 74 72 61 64 69 74 | 69 6f 6e 61 6c 20 73 74 |a tradit|ional st|
|00000190| 61 74 69 73 74 69 63 61 | 6c 20 6d 6f 64 65 6c 69 |atistica|l modeli|
|000001a0| 6e 67 20 6d 65 74 68 6f | 64 2e 20 49 74 20 69 73 |ng metho|d. It is|
|000001b0| 20 61 6e 20 69 6e 74 65 | 72 64 69 73 63 69 70 6c | an inte|rdiscipl|
|000001c0| 69 6e 61 72 79 20 61 70 | 70 72 6f 61 63 68 20 74 |inary ap|proach t|
|000001d0| 6f 20 6f 76 65 72 63 6f | 6d 65 20 73 6f 6d 65 20 |o overco|me some |
|000001e0| 6d 61 69 6e 20 64 69 73 | 61 64 76 61 6e 74 61 67 |main dis|advantag|
|000001f0| 65 73 20 6f 66 20 73 74 | 61 74 69 73 74 69 63 73 |es of st|atistics|
|00000200| 20 61 6e 64 20 4e 4e 27 | 73 2e 20 42 65 6c 6f 77 | and NN'|s. Below|
|00000210| 20 69 73 20 61 20 64 65 | 73 63 72 69 70 74 69 6f | is a de|scriptio|
|00000220| 6e 20 6f 66 20 47 4d 44 | 48 20 66 72 6f 6d 20 74 |n of GMD|H from t|
|00000230| 68 65 20 70 72 65 66 61 | 63 65 20 74 6f 20 46 61 |he prefa|ce to Fa|
|00000240| 72 6c 6f 77 27 73 20 42 | 6f 6f 6b 2e 0d 0d 50 72 |rlow's B|ook...Pr|
|00000250| 65 66 61 63 65 20 0d 49 | 6e 20 73 74 61 74 69 73 |eface .I|n statis|
|00000260| 74 69 63 73 20 6e 6f 77 | 61 64 61 79 73 20 74 68 |tics now|adays th|
|00000270| 65 72 65 20 69 73 20 61 | 20 64 69 73 74 69 6e 67 |ere is a| disting|
|00000280| 75 69 73 68 61 62 6c 65 | 20 74 72 65 6e 64 20 61 |uishable| trend a|
|00000290| 77 61 79 20 66 72 6f 6d | 20 74 68 65 20 72 65 73 |way from| the res|
|000002a0| 74 72 69 63 74 69 76 65 | 20 61 73 73 75 6d 70 74 |trictive| assumpt|
|000002b0| 69 6f 6e 73 20 6f 66 20 | 70 61 72 61 6d 65 74 72 |ions of |parametr|
|000002c0| 69 63 20 61 6e 61 6c 79 | 73 69 73 20 61 6e 64 20 |ic analy|sis and |
|000002d0| 74 6f 77 61 72 64 20 74 | 68 65 20 6d 6f 72 65 20 |toward t|he more |
|000002e0| 63 6f 6d 70 75 74 65 72 | 2d 6f 72 69 65 6e 74 65 |computer|-oriente|
|000002f0| 64 20 61 72 65 61 20 6f | 66 20 77 68 61 74 20 69 |d area o|f what i|
|00000300| 73 20 67 65 6e 65 72 61 | 6c 6c 79 20 6b 6e 6f 77 |s genera|lly know|
|00000310| 6e 20 61 73 20 20 6e 6f | 6e 70 61 72 61 6d 65 74 |n as no|nparamet|
|00000320| 72 69 63 20 64 61 74 61 | 20 61 6e 61 6c 79 73 69 |ric data| analysi|
|00000330| 73 2e 20 4f 6e 65 20 6f | 66 20 74 68 65 20 6d 6f |s. One o|f the mo|
|00000340| 72 65 20 66 61 73 63 69 | 6e 61 74 69 6e 67 20 63 |re fasci|nating c|
|00000350| 6f 6e 63 65 70 74 73 20 | 66 72 6f 6d 20 74 68 69 |oncepts |from thi|
|00000360| 73 20 6e 65 77 20 67 65 | 6e 65 72 61 74 69 6f 6e |s new ge|neration|
|00000370| 20 6f 66 20 72 65 73 65 | 61 72 63 68 20 69 73 20 | of rese|arch is |
|00000380| 77 68 61 74 20 69 73 20 | 6b 6e 6f 77 6e 20 61 73 |what is |known as|
|00000390| 20 74 68 65 20 47 4d 44 | 48 20 61 6c 67 6f 72 69 | the GMD|H algori|
|000003a0| 74 68 6d 2c 20 77 68 69 | 63 68 20 77 61 73 20 69 |thm, whi|ch was i|
|000003b0| 6e 74 72 6f 64 75 63 65 | 64 20 61 6e 64 20 69 73 |ntroduce|d and is|
|000003c0| 20 63 75 72 72 65 6e 74 | 6c 79 20 62 65 69 6e 67 | current|ly being|
|000003d0| 20 64 65 76 65 6c 6f 70 | 65 64 20 62 79 20 74 68 | develop|ed by th|
|000003e0| 65 20 55 6b 72 61 69 6e | 69 61 6e 20 63 79 62 65 |e Ukrain|ian cybe|
|000003f0| 72 6e 65 74 69 63 69 73 | 74 20 61 6e 64 20 65 6e |rneticis|t and en|
|00000400| 67 69 6e 65 65 72 20 41 | 2e 47 20 49 76 61 6b 68 |gineer A|.G Ivakh|
|00000410| 6e 65 6e 6b 6f 2e 20 57 | 68 61 74 20 69 73 20 6b |nenko. W|hat is k|
|00000420| 6e 6f 77 6e 20 74 68 65 | 73 65 20 64 61 79 73 20 |nown the|se days |
|00000430| 61 73 20 61 20 68 65 75 | 72 69 73 74 69 63 2c 20 |as a heu|ristic, |
|00000440| 74 68 65 20 47 4d 44 48 | 20 61 6c 67 6f 72 69 74 |the GMDH| algorit|
|00000450| 68 6d 20 63 6f 6e 73 74 | 72 75 63 74 73 20 68 69 |hm const|ructs hi|
|00000460| 67 68 2d 6f 72 64 65 72 | 20 72 65 67 72 65 73 73 |gh-order| regress|
|00000470| 69 6f 6e 2d 74 79 70 65 | 20 6d 6f 64 65 6c 73 20 |ion-type| models |
|00000480| 66 6f 72 20 63 6f 6d 70 | 6c 65 78 20 73 79 73 74 |for comp|lex syst|
|00000490| 65 6d 73 20 61 6e 64 20 | 68 61 73 20 74 68 65 20 |ems and |has the |
|000004a0| 61 64 76 61 6e 74 61 67 | 65 20 6f 76 65 72 20 74 |advantag|e over t|
|000004b0| 72 61 64 69 74 69 6f 6e | 61 6c 20 6d 6f 64 65 6c |radition|al model|
|000004c0| 69 6e 67 20 69 6e 20 74 | 68 61 74 20 74 68 65 20 |ing in t|hat the |
|000004d0| 6d 6f 64 65 6c 65 72 20 | 63 61 6e 20 6d 6f 72 65 |modeler |can more|
|000004e0| 2d 6f 72 2d 6c 65 73 73 | 20 74 68 72 6f 77 20 69 |-or-less| throw i|
|000004f0| 6e 74 6f 20 74 68 65 20 | 61 6c 67 6f 72 69 74 68 |nto the |algorith|
|00000500| 6d 20 61 6c 6c 20 73 6f | 72 74 73 20 6f 66 20 69 |m all so|rts of i|
|00000510| 6e 70 75 74 2f 20 6f 75 | 74 70 75 74 20 74 79 70 |nput/ ou|tput typ|
|00000520| 65 73 20 6f 66 20 6f 62 | 73 65 72 76 61 74 69 6f |es of ob|servatio|
|00000530| 6e 73 2c 20 61 6e 64 20 | 74 68 65 20 63 6f 6d 70 |ns, and |the comp|
|00000540| 75 74 65 72 20 64 6f 65 | 73 20 74 68 65 20 72 65 |uter doe|s the re|
|00000550| 73 74 2e 20 54 68 65 20 | 63 6f 6d 70 75 74 65 72 |st. The |computer|
|00000560| 20 73 65 6c 66 2d 6f 72 | 67 61 6e 69 7a 65 73 20 | self-or|ganizes |
|00000570| 74 68 65 20 6d 6f 64 65 | 6c 20 66 72 6f 6d 20 61 |the mode|l from a|
|00000580| 20 73 69 6d 70 6c 65 20 | 6f 6e 65 20 74 6f 20 6f | simple |one to o|
|00000590| 6e 65 20 6f 66 20 6f 70 | 74 69 6d 61 6c 20 63 6f |ne of op|timal co|
|000005a0| 6d 70 6c 65 78 69 74 79 | 20 62 79 20 61 20 6d 65 |mplexity| by a me|
|000005b0| 74 68 6f 64 6f 6c 6f 67 | 79 20 6e 6f 74 20 75 6e |thodolog|y not un|
|000005c0| 6c 69 6b 65 20 74 68 65 | 20 70 72 6f 63 65 73 73 |like the| process|
|000005d0| 20 6f 66 20 6e 61 74 75 | 72 61 6c 20 65 76 6f 6c | of natu|ral evol|
|000005e0| 75 74 69 6f 6e 2e 20 49 | 74 20 69 73 20 74 68 65 |ution. I|t is the|
|000005f0| 20 70 75 72 70 6f 73 65 | 20 6f 66 20 74 68 69 73 | purpose| of this|
|00000600| 20 62 6f 6f 6b 20 74 6f | 20 69 6e 74 72 6f 64 75 | book to| introdu|
|00000610| 63 65 20 74 6f 20 45 6e | 67 6c 69 73 68 2d 73 70 |ce to En|glish-sp|
|00000620| 65 61 6b 69 6e 67 20 70 | 65 6f 70 6c 65 20 74 68 |eaking p|eople th|
|00000630| 65 20 62 61 73 69 63 20 | 47 4d 44 48 20 61 6c 67 |e basic |GMDH alg|
|00000640| 6f 72 69 74 68 6d 2c 20 | 70 72 65 73 65 6e 74 20 |orithm, |present |
|00000650| 76 61 72 69 61 74 69 6f | 6e 73 20 61 6e 64 20 65 |variatio|ns and e|
|00000660| 78 61 6d 70 6c 65 73 20 | 6f 66 20 69 74 73 20 75 |xamples |of its u|
|00000670| 73 65 20 61 6e 64 20 6c | 69 73 74 20 61 20 62 69 |se and l|ist a bi|
|00000680| 62 6c 69 6f 67 72 61 70 | 68 79 20 6f 66 20 61 6c |bliograp|hy of al|
|00000690| 6c 20 70 75 62 6c 69 73 | 68 65 64 20 77 6f 72 6b |l publis|hed work|
|000006a0| 20 69 6e 20 74 68 69 73 | 20 67 72 6f 77 69 6e 67 | in this| growing|
|000006b0| 20 61 72 65 61 20 6f 66 | 20 72 65 73 65 61 72 63 | area of| researc|
|000006c0| 68 2e 20 0d 0d 22 46 6f | 72 74 79 20 79 65 61 72 |h. .."Fo|rty year|
|000006d0| 73 20 61 67 6f 20 73 63 | 69 65 6e 74 69 73 74 73 |s ago sc|ientists|
|000006e0| 20 62 65 67 61 6e 20 73 | 65 65 6b 69 6e 67 20 73 | began s|eeking s|
|000006f0| 75 63 68 20 69 6e 64 75 | 63 74 69 76 65 20 61 6c |uch indu|ctive al|
|00000700| 67 6f 72 69 74 68 6d 73 | 20 69 6e 20 74 68 65 69 |gorithms| in thei|
|00000710| 72 20 71 75 65 73 74 20 | 66 6f 72 20 75 6e 64 65 |r quest |for unde|
|00000720| 72 6c 79 69 6e 67 20 70 | 72 69 6e 63 69 70 6c 65 |rlying p|rinciple|
|00000730| 73 20 67 6f 76 65 72 6e | 69 6e 67 20 74 68 65 20 |s govern|ing the |
|00000740| 61 63 74 69 76 69 74 79 | 20 6f 66 20 74 68 65 20 |activity| of the |
|00000750| 63 65 6e 74 72 61 6c 20 | 6e 65 72 76 6f 75 73 20 |central |nervous |
|00000760| 73 79 73 74 65 6d 2e 20 | 49 74 20 77 61 73 20 62 |system. |It was b|
|00000770| 65 6c 69 65 76 65 64 20 | 2d 20 77 69 74 68 20 67 |elieved |- with g|
|00000780| 6f 6f 64 20 72 65 61 73 | 6f 6e 20 2d 20 74 68 61 |ood reas|on - tha|
|00000790| 74 20 61 20 67 72 61 73 | 70 20 6f 66 20 74 68 65 |t a gras|p of the|
|000007a0| 73 65 20 70 72 69 6e 63 | 69 70 6c 65 73 20 77 6f |se princ|iples wo|
|000007b0| 75 6c 64 20 79 69 65 6c | 64 20 69 6d 70 72 6f 76 |uld yiel|d improv|
|000007c0| 65 6d 65 6e 74 73 20 69 | 6e 20 66 65 65 64 62 61 |ements i|n feedba|
|000007d0| 63 6b 20 63 6f 6e 74 72 | 6f 6c 20 73 79 73 74 65 |ck contr|ol syste|
|000007e0| 6d 73 20 61 6e 64 20 69 | 6e 20 74 68 65 20 64 65 |ms and i|n the de|
|000007f0| 73 69 67 6e 20 6f 66 20 | 61 75 74 6f 6d 61 74 69 |sign of |automati|
|00000800| 63 20 63 61 6c 63 75 6c | 61 74 69 6e 67 20 6d 61 |c calcul|ating ma|
|00000810| 63 68 69 6e 65 73 2e 20 | 49 6e 20 74 68 65 20 55 |chines. |In the U|
|00000820| 6e 69 74 65 64 20 53 74 | 61 74 65 73 2c 20 61 6c |nited St|ates, al|
|00000830| 67 6f 72 69 74 68 6d 73 | 20 72 65 6c 61 74 65 64 |gorithms| related|
|00000840| 20 74 6f 20 47 4d 44 48 | 20 61 72 65 20 74 72 61 | to GMDH| are tra|
|00000850| 63 65 61 62 6c 65 20 64 | 69 72 65 63 74 6c 79 20 |ceable d|irectly |
|00000860| 74 6f 20 73 74 75 64 69 | 65 73 20 69 6e 20 74 68 |to studi|es in th|
|00000870| 65 20 31 39 34 30 73 20 | 6f 66 20 74 68 65 20 62 |e 1940s |of the b|
|00000880| 65 68 61 76 69 6f 75 72 | 20 6f 66 20 6e 65 75 72 |ehaviour| of neur|
|00000890| 6f 6e 73 20 61 6e 64 20 | 6e 65 75 72 6f 6e 20 61 |ons and |neuron a|
|000008a0| 67 67 72 65 67 61 74 65 | 73 2e 20 49 6e 20 74 68 |ggregate|s. In th|
|000008b0| 65 20 55 53 53 52 2c 20 | 6d 6f 72 65 20 65 6d 70 |e USSR, |more emp|
|000008c0| 68 61 73 69 73 20 77 61 | 73 20 64 69 72 65 63 74 |hasis wa|s direct|
|000008d0| 65 64 20 74 6f 77 61 72 | 64 20 74 68 65 20 6d 61 |ed towar|d the ma|
|000008e0| 74 68 65 6d 61 74 69 63 | 73 20 6f 66 20 63 79 62 |thematic|s of cyb|
|000008f0| 65 72 6e 65 74 69 63 20 | 73 79 73 74 65 6d 73 20 |ernetic |systems |
|00000900| 74 68 61 6e 20 74 6f 77 | 61 72 64 20 74 68 65 20 |than tow|ard the |
|00000910| 65 6d 75 6c 61 74 69 6f | 6e 20 6f 66 20 6e 65 75 |emulatio|n of neu|
|00000920| 72 6f 6e 73 2e 20 42 6f | 74 68 20 6c 69 6e 65 73 |rons. Bo|th lines|
|00000930| 20 6f 66 20 64 65 76 65 | 6c 6f 70 6d 65 6e 74 20 | of deve|lopment |
|00000940| 62 65 67 61 6e 20 74 6f | 20 66 6c 6f 77 20 74 6f |began to| flow to|
|00000950| 67 65 74 68 65 72 20 66 | 6f 72 20 74 68 65 20 31 |gether f|or the 1|
|00000960| 39 37 30 73 2c 20 61 6e | 64 20 69 6e 20 74 68 65 |970s, an|d in the|
|00000970| 20 70 61 73 74 20 66 69 | 76 65 20 79 65 61 72 73 | past fi|ve years|
|00000980| 2c 20 73 63 69 65 6e 74 | 69 73 74 73 20 69 6e 20 |, scient|ists in |
|00000990| 4a 61 70 61 6e 20 68 61 | 76 65 20 61 6c 73 6f 20 |Japan ha|ve also |
|000009a0| 68 61 64 20 61 6e 20 69 | 6d 70 61 63 74 20 6f 6e |had an i|mpact on|
|000009b0| 20 74 68 65 20 63 6f 75 | 72 73 65 20 6f 66 20 47 | the cou|rse of G|
|000009c0| 4d 44 48 20 77 6f 72 6b | 2e 0d 20 20 20 20 20 20 |MDH work|.. |
|000009d0| 20 20 20 20 20 20 20 20 | 20 20 20 20 20 20 20 20 | | |
|000009e0| 20 20 20 20 20 20 20 20 | 20 20 20 20 20 20 20 20 | | |
|000009f0| 20 20 20 20 20 20 20 20 | 20 20 20 20 20 20 20 20 | | |
|00000a00| 20 20 53 2e 20 4a 2e 20 | 46 61 72 6c 6f 77 2c 20 | S. J. |Farlow, |
|00000a10| 53 65 6c 66 2d 4f 72 67 | 61 6e 69 7a 69 6e 67 20 |Self-Org|anizing |
|00000a20| 6d 65 74 68 6f 64 73 20 | 69 6e 20 4d 6f 64 65 6c |methods |in Model|
|00000a30| 69 6e 67 2e 20 47 4d 44 | 48 20 54 79 70 65 20 41 |ing. GMD|H Type A|
|00000a40| 6c 67 6f 72 69 74 68 6d | 20 28 31 39 38 34 29 0d |lgorithm| (1984).|
|00000a50| 0d 59 6f 75 20 63 61 6e | 20 66 69 6e 64 20 61 20 |.You can| find a |
|00000a60| 73 68 6f 72 74 20 69 6e | 74 72 6f 20 69 6e 20 50 |short in|tro in P|
|00000a70| 61 70 65 72 32 20 28 73 | 65 63 74 69 6f 6e 20 53 |aper2 (s|ection S|
|00000a80| 65 6c 66 2d 6f 72 67 61 | 6e 69 7a 69 6e 67 20 6d |elf-orga|nizing m|
|00000a90| 6f 64 65 6c 69 6e 67 20 | 74 65 63 68 6e 6f 6c 6f |odeling |technolo|
|00000aa0| 67 69 65 73 29 20 6f 6e | 20 74 68 65 20 4b 6e 6f |gies) on| the Kno|
|00000ab0| 77 6c 65 64 67 65 4d 69 | 6e 65 72 20 77 65 62 20 |wledgeMi|ner web |
|00000ac0| 73 69 74 65 2e 20 59 6f | 75 20 6d 61 79 20 61 6c |site. Yo|u may al|
|00000ad0| 73 6f 20 77 61 6e 74 20 | 74 6f 20 6c 6f 6f 6b 20 |so want |to look |
|00000ae0| 61 74 20 74 68 65 20 70 | 75 62 6c 69 63 61 74 69 |at the p|ublicati|
|00000af0| 6f 6e 73 20 61 72 65 61 | 20 66 6f 72 20 6d 6f 72 |ons area| for mor|
|00000b00| 65 20 69 6e 66 6f 72 6d | 61 74 69 6f 6e 2e 0d 0d |e inform|ation...|
|00000b10| 51 3a 20 57 68 61 74 20 | 6b 69 6e 64 20 6f 66 20 |Q: What |kind of |
|00000b20| 6d 6f 64 65 6c 73 20 64 | 6f 65 73 20 4b 6e 6f 77 |models d|oes Know|
|00000b30| 6c 65 64 67 65 4d 69 6e | 65 72 20 63 72 65 61 74 |ledgeMin|er creat|
|00000b40| 65 3f 20 49 27 6d 20 6e | 6f 74 20 61 20 73 74 61 |e? I'm n|ot a sta|
|00000b50| 74 69 73 74 69 63 69 61 | 6e 20 61 6e 64 20 6e 65 |tisticia|n and ne|
|00000b60| 65 64 20 61 6e 20 6f 76 | 65 72 76 69 65 77 20 6f |ed an ov|erview o|
|00000b70| 66 20 70 6f 73 73 69 62 | 69 6c 69 74 69 65 73 20 |f possib|ilities |
|00000b80| 73 6f 20 69 20 63 61 6e | 20 61 70 70 6c 79 20 69 |so i can| apply i|
|00000b90| 74 20 74 6f 20 6d 79 20 | 66 69 65 6c 64 2e 20 0d |t to my |field. .|
|00000ba0| 41 3a 20 4b 6e 6f 77 6c | 65 64 67 65 4d 69 6e 65 |A: Knowl|edgeMine|
|00000bb0| 72 20 63 61 6e 20 63 72 | 65 61 74 65 20 74 77 6f |r can cr|eate two|
|00000bc0| 20 74 79 70 65 73 20 6f | 66 20 6d 6f 64 65 6c 73 | types o|f models|
|00000bd0| 20 61 75 74 6f 6d 61 74 | 69 63 61 6c 6c 79 3a 0d | automat|ically:.|
|00000be0| 20 20 2d 20 70 61 72 61 | 6d 65 74 72 69 63 20 72 | - para|metric r|
|00000bf0| 65 67 72 65 73 73 69 6f | 6e 20 6d 6f 64 65 6c 73 |egressio|n models|
|00000c00| 20 75 73 69 6e 67 20 47 | 4d 44 48 20 28 61 6c 6c | using G|MDH (all|
|00000c10| 20 76 65 72 73 69 6f 6e | 73 29 20 61 6e 64 0d 20 | version|s) and. |
|00000c20| 20 2d 20 6e 6f 6e 70 61 | 72 61 6d 65 74 72 69 63 | - nonpa|rametric|
|00000c30| 20 6d 6f 64 65 6c 73 20 | 28 70 61 74 74 65 72 6e | models |(pattern|
|00000c40| 73 29 20 75 73 69 6e 67 | 20 41 6e 61 6c 6f 67 20 |s) using| Analog |
|00000c50| 43 6f 6d 70 6c 65 78 69 | 6e 67 20 28 4b 4d 20 70 |Complexi|ng (KM p|
|00000c60| 72 6f 20 6f 6e 6c 79 29 | 2e 0d 0d 50 61 72 61 6d |ro only)|...Param|
|00000c70| 65 74 72 69 63 20 52 65 | 67 72 65 73 73 69 6f 6e |etric Re|gression|
|00000c80| 20 6d 6f 64 65 6c 73 0d | 47 4d 44 48 20 61 73 20 | models.|GMDH as |
|00000c90| 61 20 73 74 61 74 69 73 | 74 69 63 61 6c 20 6c 65 |a statis|tical le|
|00000ca0| 61 72 6e 69 6e 67 20 6e | 65 74 77 6f 72 6b 20 74 |arning n|etwork t|
|00000cb0| 65 63 68 6e 6f 6c 6f 67 | 79 20 73 75 62 64 69 76 |echnolog|y subdiv|
|00000cc0| 69 64 65 73 20 61 20 63 | 6f 6d 70 6c 65 78 20 6d |ides a c|omplex m|
|00000cd0| 6f 64 65 6c 69 6e 67 20 | 70 72 6f 62 6c 65 6d 20 |odeling |problem |
|00000ce0| 69 6e 74 6f 20 73 6d 61 | 6c 6c 65 72 2c 20 6d 6f |into sma|ller, mo|
|00000cf0| 72 65 20 65 61 73 69 6c | 79 20 68 61 6e 64 6c 65 |re easil|y handle|
|00000d00| 64 20 70 72 6f 62 6c 65 | 6d 73 20 28 6e 65 74 77 |d proble|ms (netw|
|00000d10| 6f 72 6b 20 6f 66 20 73 | 6f 2d 63 61 6c 6c 65 64 |ork of s|o-called|
|00000d20| 20 6e 65 75 72 6f 6e 73 | 29 20 61 6e 64 20 73 6f | neurons|) and so|
|00000d30| 6c 76 65 73 20 74 68 65 | 73 65 20 73 6d 61 6c 6c |lves the|se small|
|00000d40| 65 72 20 70 72 6f 62 6c | 65 6d 73 20 62 79 20 61 |er probl|ems by a|
|00000d50| 64 76 61 6e 63 65 64 20 | 73 74 61 74 69 73 74 69 |dvanced |statisti|
|00000d60| 63 61 6c 20 6d 65 74 68 | 6f 64 73 20 28 73 74 61 |cal meth|ods (sta|
|00000d70| 74 69 73 74 69 63 61 6c | 20 6c 65 61 72 6e 69 6e |tistical| learnin|
|00000d80| 67 29 2e 20 54 6f 20 6d | 61 6b 65 20 74 68 69 73 |g). To m|ake this|
|00000d90| 20 70 72 6f 63 65 73 73 | 20 68 69 67 68 6c 79 20 | process| highly |
|00000da0| 61 75 74 6f 6d 61 74 65 | 64 20 28 6c 65 61 72 6e |automate|d (learn|
|00000db0| 69 6e 67 20 75 6e 6b 6e | 6f 77 6e 20 72 65 6c 61 |ing unkn|own rela|
|00000dc0| 74 69 6f 6e 73 68 69 70 | 73 20 62 65 74 77 65 65 |tionship|s betwee|
|00000dd0| 6e 20 76 61 72 69 61 62 | 6c 65 73 29 20 61 6e 64 |n variab|les) and|
|00000de0| 20 74 6f 20 65 6e 73 75 | 72 65 20 72 65 6c 69 61 | to ensu|re relia|
|00000df0| 62 6c 65 20 6d 6f 64 65 | 6c 73 20 73 65 76 65 72 |ble mode|ls sever|
|00000e00| 61 6c 20 63 79 62 65 72 | 6e 65 74 69 63 20 70 72 |al cyber|netic pr|
|00000e10| 69 6e 63 69 70 6c 65 73 | 20 61 72 65 20 63 6f 6e |inciples| are con|
|00000e20| 73 69 64 65 72 65 64 20 | 61 6c 73 6f 2e 20 53 74 |sidered |also. St|
|00000e30| 61 74 69 73 74 69 63 61 | 6c 20 6c 65 61 72 6e 69 |atistica|l learni|
|00000e40| 6e 67 20 61 6e 64 20 69 | 6d 70 6c 65 6d 65 6e 74 |ng and i|mplement|
|00000e50| 61 74 69 6f 6e 20 6f 66 | 20 69 6d 70 6f 72 74 61 |ation of| importa|
|00000e60| 6e 74 20 63 79 62 65 72 | 6e 65 74 69 63 20 70 72 |nt cyber|netic pr|
|00000e70| 69 6e 63 69 70 6c 65 73 | 20 61 72 65 20 77 68 61 |inciples| are wha|
|00000e80| 74 20 6d 61 6b 65 73 20 | 47 4d 44 48 20 73 75 70 |t makes |GMDH sup|
|00000e90| 65 72 69 6f 72 20 74 6f | 20 4e 65 75 72 61 6c 20 |erior to| Neural |
|00000ea0| 4e 65 74 77 6f 72 6b 73 | 20 77 68 69 6c 65 20 74 |Networks| while t|
|00000eb0| 68 65 20 6e 65 74 77 6f | 72 6b 20 63 6f 6e 63 65 |he netwo|rk conce|
|00000ec0| 70 74 20 61 6e 64 20 63 | 79 62 65 72 6e 65 74 69 |pt and c|yberneti|
|00000ed0| 63 20 70 72 69 6e 63 69 | 70 6c 65 73 20 61 67 61 |c princi|ples aga|
|00000ee0| 69 6e 20 61 72 65 20 77 | 68 61 74 20 6d 61 6b 65 |in are w|hat make|
|00000ef0| 73 20 69 74 20 73 75 70 | 65 72 69 6f 72 20 74 6f |s it sup|erior to|
|00000f00| 20 73 74 61 74 69 73 74 | 69 63 73 2e 20 48 6f 77 | statist|ics. How|
|00000f10| 65 76 65 72 2c 20 72 65 | 73 75 6c 74 20 6f 66 20 |ever, re|sult of |
|00000f20| 61 6c 6c 20 74 68 72 65 | 65 20 6d 6f 64 65 6c 69 |all thre|e modeli|
|00000f30| 6e 67 20 6d 65 74 68 6f | 64 73 20 61 72 65 20 73 |ng metho|ds are s|
|00000f40| 74 61 74 69 73 74 69 63 | 61 6c 20 72 65 67 72 65 |tatistic|al regre|
|00000f50| 73 73 69 6f 6e 20 6d 6f | 64 65 6c 73 2e 20 54 68 |ssion mo|dels. Th|
|00000f60| 65 72 65 66 6f 72 65 2c | 20 77 65 20 61 72 65 20 |erefore,| we are |
|00000f70| 6e 6f 77 2c 20 66 6f 72 | 20 74 68 65 20 6d 6f 6d |now, for| the mom|
|00000f80| 65 6e 74 2c 20 69 6e 20 | 74 68 65 20 64 6f 6d 61 |ent, in |the doma|
|00000f90| 69 6e 20 6f 66 20 73 74 | 61 74 69 73 74 69 63 73 |in of st|atistics|
|00000fa0| 2e 0d 41 67 61 69 6e 2c | 20 77 65 20 6d 75 73 74 |..Again,| we must|
|00000fb0| 20 64 69 73 74 69 6e 67 | 75 69 73 68 20 62 65 74 | disting|uish bet|
|00000fc0| 77 65 65 6e 0d 20 20 2d | 20 73 74 61 74 69 63 20 |ween. -| static |
|00000fd0| 61 6e 64 20 64 79 6e 61 | 6d 69 63 20 6d 6f 64 65 |and dyna|mic mode|
|00000fe0| 6c 73 2c 0d 20 20 2d 20 | 74 69 6d 65 20 73 65 72 |ls,. - |time ser|
|00000ff0| 69 65 73 20 61 6e 64 20 | 69 6e 70 75 74 2d 6f 75 |ies and |input-ou|
|00001000| 74 70 75 74 20 6d 6f 64 | 65 6c 73 20 61 6e 64 0d |tput mod|els and.|
|00001010| 20 20 2d 20 6c 69 6e 65 | 61 72 20 61 6e 64 20 6e | - line|ar and n|
|00001020| 6f 6e 6c 69 6e 65 61 72 | 20 6d 6f 64 65 6c 73 2e |onlinear| models.|
|00001030| 0d 0d 41 20 73 74 61 74 | 69 63 20 6d 6f 64 65 6c |..A stat|ic model|
|00001040| 20 64 65 73 63 72 69 62 | 65 73 20 73 74 61 74 69 | describ|es stati|
|00001050| 63 20 72 65 6c 61 74 69 | 6f 6e 73 68 69 70 73 20 |c relati|onships |
|00001060| 28 6e 6f 20 64 65 70 65 | 6e 64 65 6e 63 65 20 66 |(no depe|ndence f|
|00001070| 72 6f 6d 20 74 69 6d 65 | 29 20 62 65 74 77 65 65 |rom time|) betwee|
|00001080| 6e 20 69 6e 70 75 74 20 | 76 61 72 69 61 62 6c 65 |n input |variable|
|00001090| 73 20 58 6d 20 61 6e 64 | 20 61 6e 20 6f 75 74 70 |s Xm and| an outp|
|000010a0| 75 74 20 76 61 72 69 61 | 62 6c 65 20 59 20 69 6e |ut varia|ble Y in|
|000010b0| 20 74 68 65 20 66 6f 72 | 6d 3a 0d 20 20 20 59 3d | the for|m:. Y=|
|000010c0| 66 28 58 6d 29 20 2c 20 | 6d 3d 31 2c 20 32 2c 20 |f(Xm) , |m=1, 2, |
|000010d0| 2e 2e 2e 2c 20 70 2e 0d | 46 6f 72 20 65 78 61 6d |..., p..|For exam|
|000010e0| 70 6c 65 2c 20 74 68 65 | 20 72 65 6c 61 74 69 6f |ple, the| relatio|
|000010f0| 6e 0d 20 20 20 50 52 4f | 46 49 54 20 3d 20 54 55 |n. PRO|FIT = TU|
|00001100| 52 4e 4f 56 45 52 20 2d | 20 43 4f 53 54 53 0d 69 |RNOVER -| COSTS.i|
|00001110| 73 20 61 20 73 74 61 74 | 69 63 20 6d 6f 64 65 6c |s a stat|ic model|
|00001120| 20 6f 66 20 61 6e 20 65 | 63 6f 6e 6f 6d 69 63 61 | of an e|conomica|
|00001130| 6c 20 73 79 73 74 65 6d | 2e 0d 44 79 6e 61 6d 69 |l system|..Dynami|
|00001140| 63 20 6d 6f 64 65 6c 73 | 20 61 72 65 20 75 73 65 |c models| are use|
|00001150| 64 20 74 6f 20 6d 6f 64 | 65 6c 20 61 6e 64 20 70 |d to mod|el and p|
|00001160| 72 65 64 69 63 74 20 74 | 68 65 20 64 79 6e 61 6d |redict t|he dynam|
|00001170| 69 63 20 62 65 68 61 76 | 69 6f 72 20 6f 66 20 61 |ic behav|ior of a|
|00001180| 20 74 69 6d 65 20 70 72 | 6f 63 65 73 73 2c 20 74 | time pr|ocess, t|
|00001190| 68 65 20 65 76 6f 6c 75 | 74 69 6f 6e 20 6f 66 20 |he evolu|tion of |
|000011a0| 61 20 76 61 72 69 61 62 | 6c 65 20 6f 76 65 72 20 |a variab|le over |
|000011b0| 74 69 6d 65 20 74 2e 20 | 54 68 65 72 65 66 6f 72 |time t. |Therefor|
|000011c0| 65 2c 20 61 64 64 69 74 | 69 6f 6e 61 6c 6c 79 20 |e, addit|ionally |
|000011d0| 6c 61 67 67 65 64 20 73 | 61 6d 70 6c 65 73 20 6f |lagged s|amples o|
|000011e0| 66 20 74 68 65 20 6f 75 | 74 70 75 74 20 76 61 72 |f the ou|tput var|
|000011f0| 69 61 62 6c 65 20 59 28 | 74 2d 6e 29 20 61 6e 64 |iable Y(|t-n) and|
|00001200| 2f 20 6f 72 20 6c 61 67 | 67 65 64 20 73 61 6d 70 |/ or lag|ged samp|
|00001210| 6c 65 73 20 6f 66 20 74 | 68 65 20 69 6e 70 75 74 |les of t|he input|
|00001220| 20 76 61 72 69 61 62 6c | 65 73 20 58 6d 28 74 2d | variabl|es Xm(t-|
|00001230| 6e 29 20 61 72 65 20 63 | 6f 6e 73 69 64 65 72 65 |n) are c|onsidere|
|00001240| 64 3a 0d 20 20 20 20 59 | 28 74 29 3d 66 28 58 6d |d:. Y|(t)=f(Xm|
|00001250| 28 74 29 2c 20 59 28 74 | 2d 6e 29 2c 20 58 6d 28 |(t), Y(t|-n), Xm(|
|00001260| 74 2d 6e 29 29 2c 20 6d | 3d 31 2c 20 32 2c 20 2e |t-n)), m|=1, 2, .|
|00001270| 2e 2e 2c 20 70 20 3b 20 | 6e 3d 31 2c 20 32 2c 20 |.., p ; |n=1, 2, |
|00001280| 2e 2e 2e 2c 20 71 20 2c | 0d 77 68 65 72 65 20 74 |..., q ,|.where t|
|00001290| 20 72 65 66 6c 65 63 74 | 73 20 61 20 73 70 65 63 | reflect|s a spec|
|000012a0| 69 66 69 63 20 74 69 6d | 65 20 68 6f 72 69 7a 6f |ific tim|e horizo|
|000012b0| 6e 20 28 64 61 79 2c 20 | 77 65 65 6b 2c 20 6d 6f |n (day, |week, mo|
|000012c0| 6e 74 68 20 65 74 63 29 | 2e 0d 46 6f 72 20 65 78 |nth etc)|..For ex|
|000012d0| 61 6d 70 6c 65 2c 20 74 | 68 65 20 72 65 6c 61 74 |ample, t|he relat|
|000012e0| 69 6f 6e 0d 20 20 20 50 | 52 4f 46 49 54 28 74 29 |ion. P|ROFIT(t)|
|000012f0| 20 3d 20 31 2e 31 20 2a | 20 50 52 4f 46 49 54 28 | = 1.1 *| PROFIT(|
|00001300| 74 2d 31 29 0d 69 73 20 | 61 20 73 69 6d 70 6c 65 |t-1).is |a simple|
|00001310| 20 64 79 6e 61 6d 69 63 | 20 6d 6f 64 65 6c 20 74 | dynamic| model t|
|00001320| 6f 20 64 65 73 63 72 69 | 62 65 20 74 68 65 20 65 |o descri|be the e|
|00001330| 76 6f 6c 75 74 69 6f 6e | 20 6f 66 20 74 68 65 20 |volution| of the |
|00001340| 70 72 6f 66 69 74 2e 0d | 0d 41 20 74 69 6d 65 20 |profit..|.A time |
|00001350| 73 65 72 69 65 73 20 6d | 6f 64 65 6c 20 28 6f 72 |series m|odel (or|
|00001360| 20 61 75 74 6f 2d 72 65 | 67 72 65 73 73 69 76 65 | auto-re|gressive|
|00001370| 20 6d 6f 64 65 6c 29 20 | 69 73 20 69 6e 74 65 6e | model) |is inten|
|00001380| 64 65 64 20 74 6f 20 72 | 65 66 6c 65 63 74 20 74 |ded to r|eflect t|
|00001390| 68 65 20 65 76 6f 6c 75 | 74 69 6f 6e 20 6f 66 20 |he evolu|tion of |
|000013a0| 61 20 76 61 72 69 61 62 | 6c 65 20 6f 76 65 72 20 |a variab|le over |
|000013b0| 74 69 6d 65 20 6d 61 74 | 68 65 6d 61 74 69 63 61 |time mat|hematica|
|000013c0| 6c 6c 79 20 62 79 20 65 | 78 63 6c 75 73 69 76 65 |lly by e|xclusive|
|000013d0| 6c 79 20 6c 6f 6f 6b 69 | 6e 67 20 61 74 20 68 69 |ly looki|ng at hi|
|000013e0| 73 74 6f 72 69 63 61 6c | 20 64 61 74 61 20 6f 66 |storical| data of|
|000013f0| 20 74 68 65 20 76 61 72 | 69 61 62 6c 65 20 69 74 | the var|iable it|
|00001400| 73 65 6c 66 2e 20 49 74 | 20 69 73 20 61 20 6d 6f |self. It| is a mo|
|00001410| 64 65 6c 20 69 6e 20 77 | 68 69 63 68 20 74 68 65 |del in w|hich the|
|00001420| 20 6f 75 74 70 75 74 20 | 76 61 72 69 61 62 6c 65 | output |variable|
|00001430| 20 59 28 74 29 20 77 69 | 6c 6c 20 62 65 20 64 65 | Y(t) wi|ll be de|
|00001440| 73 63 72 69 62 65 64 20 | 6f 6e 6c 79 20 62 79 20 |scribed |only by |
|00001450| 69 74 73 20 6c 61 67 67 | 65 64 20 73 61 6d 70 6c |its lagg|ed sampl|
|00001460| 65 73 20 59 28 74 29 3d | 66 28 59 28 74 2d 6e 29 |es Y(t)=|f(Y(t-n)|
|00001470| 29 2c 20 77 68 65 72 65 | 62 79 20 66 20 69 73 20 |), where|by f is |
|00001480| 74 68 65 20 75 6e 6b 6e | 6f 77 6e 20 74 72 61 6e |the unkn|own tran|
|00001490| 73 66 65 72 20 66 75 6e | 63 74 69 6f 6e 20 61 6e |sfer fun|ction an|
|000014a0| 64 20 6e 20 69 73 20 74 | 68 65 20 63 68 6f 73 65 |d n is t|he chose|
|000014b0| 6e 20 6d 61 78 2e 20 74 | 69 6d 65 20 6c 61 67 20 |n max. t|ime lag |
|000014c0| 20 20 20 28 6e 20 3e 20 | 30 29 2e 20 53 69 6e 63 | (n > |0). Sinc|
|000014d0| 65 20 72 65 61 6c 20 77 | 6f 72 6c 64 20 70 72 6f |e real w|orld pro|
|000014e0| 63 65 73 73 65 73 20 61 | 72 65 20 69 6e 66 6c 75 |cesses a|re influ|
|000014f0| 65 6e 63 65 64 20 62 79 | 20 64 69 66 66 65 72 65 |enced by| differe|
|00001500| 6e 74 20 76 61 72 69 61 | 62 6c 65 73 2c 20 74 69 |nt varia|bles, ti|
|00001510| 6d 65 20 73 65 72 69 65 | 73 20 6d 6f 64 65 6c 73 |me serie|s models|
|00001520| 20 63 61 6e 20 6f 6e 6c | 79 20 68 61 76 65 20 6c | can onl|y have l|
|00001530| 69 6d 69 74 65 64 20 73 | 69 67 6e 69 66 69 63 61 |imited s|ignifica|
|00001540| 6e 63 65 20 66 6f 72 20 | 6d 6f 64 65 6c 69 6e 67 |nce for |modeling|
|00001550| 2c 20 70 72 65 64 69 63 | 74 69 6f 6e 20 61 6e 64 |, predic|tion and|
|00001560| 20 61 6e 61 6c 79 73 69 | 73 20 6f 66 20 63 6f 6d | analysi|s of com|
|00001570| 70 6c 65 78 20 70 72 6f | 63 65 73 73 65 73 2e 20 |plex pro|cesses. |
|00001580| 54 69 6d 65 20 73 65 72 | 69 65 73 20 6d 6f 64 65 |Time ser|ies mode|
|00001590| 6c 73 20 61 72 65 20 61 | 6c 77 61 79 73 20 64 79 |ls are a|lways dy|
|000015a0| 6e 61 6d 69 63 20 6d 6f | 64 65 6c 73 2e 0d 49 6e |namic mo|dels..In|
|000015b0| 70 75 74 2d 6f 75 74 70 | 75 74 20 6d 6f 64 65 6c |put-outp|ut model|
|000015c0| 73 20 63 61 6e 20 65 69 | 74 68 65 72 20 62 65 20 |s can ei|ther be |
|000015d0| 73 74 61 74 69 63 61 6c | 6c 79 20 6f 72 20 64 79 |statical|ly or dy|
|000015e0| 6e 61 6d 69 63 61 6c 6c | 79 2e 20 54 68 65 20 6f |namicall|y. The o|
|000015f0| 75 74 70 75 74 20 76 61 | 72 69 61 62 6c 65 20 59 |utput va|riable Y|
|00001600| 20 69 73 20 64 65 73 63 | 72 69 62 65 64 20 62 79 | is desc|ribed by|
|00001610| 20 61 20 6e 75 6d 62 65 | 72 20 6f 66 20 64 69 66 | a numbe|r of dif|
|00001620| 66 65 72 65 6e 74 20 69 | 6e 70 75 74 20 76 61 72 |ferent i|nput var|
|00001630| 69 61 62 6c 65 73 20 58 | 6d 3a 0d 20 20 20 59 3d |iables X|m:. Y=|
|00001640| 66 28 58 6d 29 20 2c 20 | 6d 3d 31 2c 20 32 2c 20 |f(Xm) , |m=1, 2, |
|00001650| 2e 2e 2e 2c 20 70 2e 0d | 49 6e 20 74 68 69 73 20 |..., p..|In this |
|00001660| 77 61 79 20 69 74 20 69 | 73 20 70 6f 73 73 69 62 |way it i|s possib|
|00001670| 6c 65 20 74 6f 20 66 69 | 6e 64 20 61 20 72 65 6c |le to fi|nd a rel|
|00001680| 61 74 69 6f 6e 73 68 69 | 70 20 62 65 74 77 65 65 |ationshi|p betwee|
|00001690| 6e 20 61 20 6e 75 6d 62 | 65 72 20 6f 66 20 70 6f |n a numb|er of po|
|000016a0| 74 65 6e 74 69 61 6c 20 | 69 6e 70 75 74 20 76 61 |tential |input va|
|000016b0| 72 69 61 62 6c 65 73 20 | 61 6e 64 20 61 6e 20 6f |riables |and an o|
|000016c0| 75 74 70 75 74 20 76 61 | 72 69 61 62 6c 65 20 28 |utput va|riable (|
|000016d0| 6d 75 6c 74 69 76 61 72 | 69 61 74 65 20 72 65 67 |multivar|iate reg|
|000016e0| 72 65 73 73 69 6f 6e 20 | 61 6e 61 6c 79 73 69 73 |ression |analysis|
|000016f0| 29 2e 0d 0d 49 6e 20 4b | 6e 6f 77 6c 65 64 67 65 |)...In K|nowledge|
|00001700| 4d 69 6e 65 72 2c 20 6c | 69 6e 65 61 72 20 61 6e |Miner, l|inear an|
|00001710| 64 20 6e 6f 6e 6c 69 6e | 65 61 72 20 6d 6f 64 65 |d nonlin|ear mode|
|00001720| 6c 73 20 61 72 65 20 70 | 61 72 61 6d 65 74 72 69 |ls are p|arametri|
|00001730| 63 2c 20 70 6f 6c 79 6e | 6f 6d 69 61 6c 20 6d 6f |c, polyn|omial mo|
|00001740| 64 65 6c 73 2e 20 54 68 | 65 20 6f 75 74 70 75 74 |dels. Th|e output|
|00001750| 20 76 61 72 69 61 62 6c | 65 20 59 20 69 73 20 64 | variabl|e Y is d|
|00001760| 65 73 63 72 69 62 65 64 | 20 62 79 20 61 20 70 6f |escribed| by a po|
|00001770| 6c 79 6e 6f 6d 69 61 6c | 20 66 75 6e 63 74 69 6f |lynomial| functio|
|00001780| 6e 20 6f 66 20 6f 6e 65 | 20 6f 72 20 6d 6f 72 65 |n of one| or more|
|00001790| 20 69 6e 70 75 74 20 76 | 61 72 69 61 62 6c 65 73 | input v|ariables|
|000017a0| 20 58 6d 20 20 20 28 6d | 20 3e 3d 20 32 29 3a 0d | Xm (m| >= 2):.|
|000017b0| 20 20 20 20 20 59 3d 66 | 28 58 31 2c 20 58 32 2c | Y=f|(X1, X2,|
|000017c0| 20 2e 2e 2e 2c 20 58 6d | 29 2e 0d 49 6e 20 6c 69 | ..., Xm|)..In li|
|000017d0| 6e 65 61 72 20 6d 6f 64 | 65 6c 73 20 6f 6e 6c 79 |near mod|els only|
|000017e0| 20 74 68 65 20 66 69 72 | 73 74 20 6f 72 64 65 72 | the fir|st order|
|000017f0| 20 70 6f 6c 79 6e 6f 6d | 69 61 6c 73 20 6f 66 20 | polynom|ials of |
|00001800| 20 74 68 65 20 69 6e 70 | 75 74 20 76 61 72 69 61 | the inp|ut varia|
|00001810| 62 6c 65 73 20 61 72 65 | 20 63 6f 6e 73 69 64 65 |bles are| conside|
|00001820| 72 65 64 2e 20 54 68 65 | 20 6c 69 6e 65 61 72 20 |red. The| linear |
|00001830| 6d 6f 64 65 6c 20 6f 66 | 20 74 77 6f 20 69 6e 70 |model of| two inp|
|00001840| 75 74 20 76 61 72 69 61 | 62 6c 65 73 20 77 6f 75 |ut varia|bles wou|
|00001850| 6c 64 20 6c 6f 6f 6b 20 | 6c 69 6b 65 20 74 68 69 |ld look |like thi|
|00001860| 73 3a 0d 20 20 20 20 20 | 59 3d 61 20 2b 20 62 2a |s:. |Y=a + b*|
|00001870| 58 31 20 2b 20 63 2a 58 | 32 2e 0d 41 2c 20 62 20 |X1 + c*X|2..A, b |
|00001880| 61 6e 64 20 63 20 61 72 | 65 20 74 68 65 20 6d 6f |and c ar|e the mo|
|00001890| 64 65 6c 20 70 61 72 61 | 6d 65 74 65 72 73 20 6f |del para|meters o|
|000018a0| 72 20 77 65 69 67 68 74 | 73 2e 20 0d 4e 6f 6e 6c |r weight|s. .Nonl|
|000018b0| 69 6e 65 61 72 20 6d 6f | 64 65 6c 73 20 61 72 65 |inear mo|dels are|
|000018c0| 20 63 6f 6d 70 6f 73 65 | 64 20 61 64 64 69 74 69 | compose|d additi|
|000018d0| 6f 6e 61 6c 6c 79 20 62 | 79 20 61 74 20 6c 65 61 |onally b|y at lea|
|000018e0| 73 74 20 6f 6e 65 20 68 | 69 67 68 65 72 20 6f 72 |st one h|igher or|
|000018f0| 64 65 72 20 70 6f 6c 79 | 6e 6f 6d 69 61 6c 2e 20 |der poly|nomial. |
|00001900| 41 20 20 6e 6f 6e 6c 69 | 6e 65 61 72 20 6d 6f 64 |A nonli|near mod|
|00001910| 65 6c 20 6f 66 20 74 77 | 6f 20 69 6e 70 75 74 20 |el of tw|o input |
|00001920| 76 61 72 69 61 62 6c 65 | 73 20 6d 61 79 20 6c 6f |variable|s may lo|
|00001930| 6f 6b 20 6c 69 6b 65 20 | 74 68 69 73 3a 0d 20 20 |ok like |this:. |
|00001940| 20 20 20 59 3d 61 20 2b | 20 62 2a 58 31 20 2b 20 | Y=a +| b*X1 + |
|00001950| 63 2a 58 31 2a 58 32 20 | 2b 20 64 2a 58 32 2a 58 |c*X1*X2 |+ d*X2*X|
|00001960| 32 2e 0d 0d 4e 6f 6e 70 | 61 72 61 6d 65 74 72 69 |2...Nonp|arametri|
|00001970| 63 20 6d 6f 64 65 6c 73 | 0d 4e 6f 6e 70 61 72 61 |c models|.Nonpara|
|00001980| 6d 65 74 72 69 63 20 6d | 6f 64 65 6c 73 20 61 72 |metric m|odels ar|
|00001990| 65 20 73 65 6c 65 63 74 | 65 64 20 66 72 6f 6d 20 |e select|ed from |
|000019a0| 61 20 67 69 76 65 6e 20 | 76 61 72 69 61 62 6c 65 |a given |variable|
|000019b0| 73 20 73 65 74 20 6f 66 | 20 6f 62 73 65 72 76 61 |s set of| observa|
|000019c0| 74 69 6f 6e 73 20 62 79 | 20 41 6e 61 6c 6f 67 20 |tions by| Analog |
|000019d0| 43 6f 6d 70 6c 65 78 69 | 6e 67 20 72 65 70 72 65 |Complexi|ng repre|
|000019e0| 73 65 6e 74 69 6e 67 20 | 6f 6e 65 20 6f 72 20 6d |senting |one or m|
|000019f0| 6f 72 65 20 70 61 74 74 | 65 72 6e 73 20 6f 66 20 |ore patt|erns of |
|00001a00| 61 20 74 72 61 6a 65 63 | 74 6f 72 79 20 6f 66 20 |a trajec|tory of |
|00001a10| 70 61 73 74 20 62 65 68 | 61 76 69 6f 72 20 77 68 |past beh|avior wh|
|00001a20| 69 63 68 20 61 72 65 20 | 61 6e 61 6c 6f 67 6f 75 |ich are |analogou|
|00001a30| 73 20 74 6f 20 61 20 63 | 68 6f 73 65 6e 20 72 65 |s to a c|hosen re|
|00001a40| 66 65 72 65 6e 63 65 20 | 70 61 74 74 65 72 6e 2e |ference |pattern.|
|00001a50| 20 54 68 65 79 20 61 72 | 65 20 63 61 6c 6c 65 64 | They ar|e called|
|00001a60| 20 6e 6f 6e 70 61 72 61 | 6d 65 74 72 69 63 20 73 | nonpara|metric s|
|00001a70| 69 6e 63 65 20 74 68 65 | 72 65 20 69 73 20 6e 6f |ince the|re is no|
|00001a80| 20 6e 65 65 64 20 74 6f | 20 65 73 74 69 6d 61 74 | need to| estimat|
|00001a90| 65 20 73 6f 6d 65 20 70 | 61 72 61 6d 65 74 65 72 |e some p|arameter|
|00001aa0| 73 2e 20 41 6e 61 6c 6f | 67 20 43 6f 6d 70 6c 65 |s. Analo|g Comple|
|00001ab0| 78 69 6e 67 20 69 73 20 | 62 61 73 65 64 20 6f 6e |xing is |based on|
|00001ac0| 20 74 68 65 20 61 73 73 | 75 6d 70 74 69 6f 6e 20 | the ass|umption |
|00001ad0| 74 68 61 74 20 74 68 65 | 72 65 20 65 78 69 73 74 |that the|re exist|
|00001ae0| 20 74 79 70 69 63 61 6c | 20 73 69 74 75 61 74 69 | typical| situati|
|00001af0| 6f 6e 73 20 6f 66 20 61 | 20 73 70 65 63 69 66 69 |ons of a| specifi|
|00001b00| 63 20 70 72 6f 63 65 73 | 73 2c 20 69 2e 65 2e 20 |c proces|s, i.e. |
|00001b10| 65 61 63 68 20 61 63 74 | 75 61 6c 20 70 65 72 69 |each act|ual peri|
|00001b20| 6f 64 20 6f 66 20 73 74 | 61 74 65 20 65 76 6f 6c |od of st|ate evol|
|00001b30| 75 74 69 6f 6e 20 6f 66 | 20 61 20 67 69 76 65 6e |ution of| a given|
|00001b40| 20 6d 75 6c 74 69 64 69 | 6d 65 6e 73 69 6f 6e 61 | multidi|mensiona|
|00001b50| 6c 20 74 69 6d 65 20 70 | 72 6f 63 65 73 73 20 6d |l time p|rocess m|
|00001b60| 61 79 20 68 61 76 65 20 | 6f 6e 65 20 6f 72 20 6d |ay have |one or m|
|00001b70| 6f 72 65 20 61 6e 61 6c | 6f 67 75 65 73 20 69 6e |ore anal|ogues in|
|00001b80| 20 68 69 73 74 6f 72 79 | 2e 20 49 66 20 73 6f 2c | history|. If so,|
|00001b90| 20 69 74 20 77 69 6c 6c | 20 62 65 20 6c 69 6b 65 | it will| be like|
|00001ba0| 6c 79 20 74 68 61 74 20 | 61 20 70 72 65 64 69 63 |ly that |a predic|
|00001bb0| 74 69 6f 6e 20 63 6f 75 | 6c 64 20 62 65 20 6f 62 |tion cou|ld be ob|
|00001bc0| 74 61 69 6e 65 64 20 62 | 79 20 74 72 61 6e 73 66 |tained b|y transf|
|00001bd0| 6f 72 6d 69 6e 67 20 74 | 68 65 20 6b 6e 6f 77 6e |orming t|he known|
|00001be0| 20 63 6f 6e 74 69 6e 75 | 61 74 69 6f 6e 73 20 6f | continu|ations o|
|00001bf0| 66 20 74 68 65 20 68 69 | 73 74 6f 72 69 63 61 6c |f the hi|storical|
|00001c00| 20 61 6e 61 6c 6f 67 75 | 65 73 2e 20 49 74 20 69 | analogu|es. It i|
|00001c10| 73 20 65 73 73 65 6e 74 | 69 61 6c 20 74 68 61 74 |s essent|ial that|
|00001c20| 20 73 65 61 72 63 68 69 | 6e 67 20 66 6f 72 20 61 | searchi|ng for a|
|00001c30| 6e 61 6c 6f 67 6f 75 73 | 20 70 61 74 74 65 72 6e |nalogous| pattern|
|00001c40| 20 69 73 20 6e 6f 74 20 | 6f 6e 6c 79 20 70 72 6f | is not |only pro|
|00001c50| 63 65 73 73 65 64 20 6f | 6e 20 61 20 73 69 6e 67 |cessed o|n a sing|
|00001c60| 6c 65 20 73 74 61 74 65 | 20 76 61 72 69 61 62 6c |le state| variabl|
|00001c70| 65 20 28 74 69 6d 65 20 | 73 65 72 69 65 73 29 20 |e (time |series) |
|00001c80| 62 75 74 20 6f 6e 20 61 | 20 73 65 74 20 6f 66 20 |but on a| set of |
|00001c90| 72 65 70 72 65 73 65 6e | 74 61 74 69 76 65 20 76 |represen|tative v|
|00001ca0| 61 72 69 61 62 6c 65 73 | 20 73 69 6d 75 6c 74 61 |ariables| simulta|
|00001cb0| 6e 65 6f 75 73 6c 79 20 | 61 6e 64 20 6f 62 6a 65 |neously |and obje|
|00001cc0| 63 74 69 76 65 6c 79 2e | 20 54 68 65 73 65 20 6e |ctively.| These n|
|00001cd0| 6f 6e 70 61 72 61 6d 65 | 74 72 69 63 20 6d 6f 64 |onparame|tric mod|
|00001ce0| 65 6c 73 20 63 61 6e 20 | 62 65 20 75 73 65 64 20 |els can |be used |
|00001cf0| 74 6f 20 6d 6f 64 65 6c | 20 61 6e 64 20 70 72 65 |to model| and pre|
|00001d00| 64 69 63 74 20 74 68 65 | 20 6d 6f 73 74 20 66 75 |dict the| most fu|
|00001d10| 7a 7a 79 20 74 69 6d 65 | 20 70 72 6f 63 65 73 73 |zzy time| process|
|00001d20| 65 73 20 6c 69 6b 65 20 | 74 68 65 20 65 76 6f 6c |es like |the evol|
|00001d30| 75 74 69 6f 6e 20 6f 66 | 20 6d 61 72 6b 65 74 73 |ution of| markets|
|00001d40| 2e 20 41 6e 61 6c 6f 67 | 20 43 6f 6d 70 6c 65 78 |. Analog| Complex|
|00001d50| 69 6e 67 20 61 70 70 6c | 69 65 64 20 74 6f 20 66 |ing appl|ied to f|
|00001d60| 69 6e 61 6e 63 69 6e 67 | 2c 20 66 6f 72 20 65 78 |inancing|, for ex|
|00001d70| 61 6d 70 6c 65 2c 20 63 | 61 6e 20 62 65 20 76 69 |ample, c|an be vi|
|00001d80| 65 77 65 64 20 61 73 20 | 61 20 6b 69 6e 64 20 6f |ewed as |a kind o|
|00001d90| 66 20 6d 75 6c 74 69 64 | 69 6d 65 6e 73 69 6f 6e |f multid|imension|
|00001da0| 61 6c 2c 20 61 75 74 6f | 6d 61 74 65 64 20 61 6e |al, auto|mated an|
|00001db0| 64 20 6f 62 6a 65 63 74 | 69 76 65 6c 79 20 77 6f |d object|ively wo|
|00001dc0| 72 6b 69 6e 67 20 43 68 | 61 72 74 20 41 6e 61 6c |rking Ch|art Anal|
|00001dd0| 79 73 69 73 28 73 65 65 | 20 50 61 70 65 72 32 20 |ysis(see| Paper2 |
|00001de0| 69 6e 20 74 68 65 20 77 | 65 62 20 73 69 74 65 20 |in the w|eb site |
|00001df0| 66 6f 72 20 61 20 70 69 | 63 74 75 72 65 29 2e 0d |for a pi|cture)..|
|00001e00| 0d 51 3a 20 49 6e 20 74 | 68 65 20 70 72 6f 67 72 |.Q: In t|he progr|
|00001e10| 61 6d 20 49 20 66 6f 75 | 6e 64 20 74 68 65 20 74 |am I fou|nd the t|
|00001e20| 65 72 6d 20 27 73 79 73 | 74 65 6d 20 6f 66 20 65 |erm 'sys|tem of e|
|00001e30| 71 75 61 74 69 6f 6e 73 | 27 2e 20 57 68 61 74 20 |quations|'. What |
|00001e40| 64 6f 65 73 20 69 74 20 | 6d 65 61 6e 20 61 6e 64 |does it |mean and|
|00001e50| 20 77 68 65 6e 20 64 6f | 20 49 20 68 61 76 65 20 | when do| I have |
|00001e60| 74 6f 20 63 72 65 61 74 | 65 20 61 20 73 79 73 74 |to creat|e a syst|
|00001e70| 65 6d 3f 0d 41 3a 20 41 | 20 73 79 73 74 65 6d 20 |em?.A: A| system |
|00001e80| 6f 66 20 65 71 75 61 74 | 69 6f 6e 73 20 63 6f 6e |of equat|ions con|
|00001e90| 73 69 73 74 73 20 6f 66 | 20 6d 75 6c 74 69 70 6c |sists of| multipl|
|00001ea0| 65 2c 20 69 6e 74 65 72 | 63 6f 6e 6e 65 63 74 65 |e, inter|connecte|
|00001eb0| 64 20 69 6e 70 75 74 2d | 6f 75 74 70 75 74 20 6d |d input-|output m|
|00001ec0| 6f 64 65 6c 73 2e 20 54 | 68 65 72 65 66 6f 72 65 |odels. T|herefore|
|00001ed0| 2c 20 69 74 20 69 73 20 | 73 6f 6d 65 74 69 6d 65 |, it is |sometime|
|00001ee0| 73 20 63 61 6c 6c 65 64 | 20 61 20 6d 75 6c 74 69 |s called| a multi|
|00001ef0| 2d 69 6e 70 75 74 2f 6d | 75 6c 74 69 2d 6f 75 74 |-input/m|ulti-out|
|00001f00| 70 75 74 20 73 79 73 74 | 65 6d 2e 20 49 74 20 69 |put syst|em. It i|
|00001f10| 73 20 70 6f 73 73 69 62 | 6c 65 20 74 6f 20 63 72 |s possib|le to cr|
|00001f20| 65 61 74 65 20 61 20 73 | 74 61 74 69 63 20 73 79 |eate a s|tatic sy|
|00001f30| 73 74 65 6d 20 28 63 6f | 6e 73 69 73 74 69 6e 67 |stem (co|nsisting|
|00001f40| 20 6f 66 20 73 74 61 74 | 69 63 20 69 6e 70 75 74 | of stat|ic input|
|00001f50| 2d 6f 75 74 70 75 74 20 | 6d 6f 64 65 6c 73 20 6f |-output |models o|
|00001f60| 6e 6c 79 29 20 6f 72 20 | 61 20 64 79 6e 61 6d 69 |nly) or |a dynami|
|00001f70| 63 20 73 79 73 74 65 6d | 20 6f 66 20 65 71 75 61 |c system| of equa|
|00001f80| 74 69 6f 6e 73 2e 20 46 | 6f 72 20 65 78 61 6d 70 |tions. F|or examp|
|00001f90| 6c 65 2c 20 61 20 64 79 | 6e 61 6d 69 63 20 73 79 |le, a dy|namic sy|
|00001fa0| 73 74 65 6d 20 6f 66 20 | 33 20 76 61 72 69 61 62 |stem of |3 variab|
|00001fb0| 6c 65 73 20 6d 61 79 20 | 6c 6f 6f 6b 20 6c 69 6b |les may |look lik|
|00001fc0| 65 20 74 68 69 73 3a 0d | 20 20 20 58 31 28 74 29 |e this:.| X1(t)|
|00001fd0| 3d 66 31 28 58 32 28 74 | 29 2c 20 58 33 28 74 29 |=f1(X2(t|), X3(t)|
|00001fe0| 2c 20 58 31 28 74 2d 31 | 29 2c 20 58 32 28 74 2d |, X1(t-1|), X2(t-|
|00001ff0| 33 29 2c 20 58 33 28 74 | 2d 35 29 29 0d 20 20 20 |3), X3(t|-5)). |
|00002000| 58 32 28 74 29 3d 66 32 | 28 58 31 28 74 29 2c 20 |X2(t)=f2|(X1(t), |
|00002010| 58 33 28 74 29 2c 20 58 | 32 28 74 2d 33 29 2c 20 |X3(t), X|2(t-3), |
|00002020| 58 31 28 74 2d 35 29 29 | 0d 20 20 20 58 33 28 74 |X1(t-5))|. X3(t|
|00002030| 29 3d 66 32 28 58 31 28 | 74 29 2c 20 58 32 28 74 |)=f2(X1(|t), X2(t|
|00002040| 29 2c 20 58 33 28 74 2d | 31 29 2c 20 58 31 28 74 |), X3(t-|1), X1(t|
|00002050| 2d 33 29 29 20 2e 0d 49 | 66 20 77 65 20 77 61 6e |-3)) ..I|f we wan|
|00002060| 74 20 6e 6f 74 20 6f 6e | 6c 79 20 74 6f 20 6d 6f |t not on|ly to mo|
|00002070| 64 65 6c 20 61 20 73 79 | 73 74 65 6d 20 62 75 74 |del a sy|stem but|
|00002080| 20 61 6c 73 6f 20 74 6f | 20 70 72 65 64 69 63 74 | also to| predict|
|00002090| 20 69 74 2c 20 77 65 20 | 6d 75 73 74 20 73 61 74 | it, we |must sat|
|000020a0| 69 73 66 79 20 61 6e 20 | 61 64 64 69 74 69 6f 6e |isfy an |addition|
|000020b0| 61 6c 20 69 6d 70 6f 72 | 74 61 6e 74 20 63 6f 6e |al impor|tant con|
|000020c0| 64 69 74 69 6f 6e 3a 20 | 69 74 20 68 61 73 20 74 |dition: |it has t|
|000020d0| 6f 20 62 65 20 63 6f 6e | 66 6c 69 63 74 20 66 72 |o be con|flict fr|
|000020e0| 65 65 2e 20 49 6e 20 74 | 68 65 20 73 79 73 74 65 |ee. In t|he syste|
|000020f0| 6d 20 61 62 6f 76 65 20 | 65 61 63 68 20 6f 75 74 |m above |each out|
|00002100| 70 75 74 20 76 61 72 69 | 61 62 6c 65 20 64 65 70 |put vari|able dep|
|00002110| 65 6e 64 73 20 6f 6e 20 | 61 6c 6c 20 6f 74 68 65 |ends on |all othe|
|00002120| 72 73 20 61 74 20 74 69 | 6d 65 20 74 2e 20 54 68 |rs at ti|me t. Th|
|00002130| 69 73 20 73 79 73 74 65 | 6d 20 69 73 20 6e 6f 74 |is syste|m is not|
|00002140| 20 66 72 65 65 20 6f 66 | 20 63 6f 6e 66 6c 69 63 | free of| conflic|
|00002150| 74 73 20 61 6e 64 20 74 | 68 65 72 65 66 6f 72 65 |ts and t|herefore|
|00002160| 20 6e 6f 74 20 61 70 70 | 6c 69 63 61 62 6c 65 20 | not app|licable |
|00002170| 66 6f 72 20 70 72 65 64 | 69 63 74 69 6f 6e 2e 20 |for pred|iction. |
|00002180| 0d 0d 4b 6e 6f 77 6c 65 | 64 67 65 4d 69 6e 65 72 |..Knowle|dgeMiner|
|00002190| 20 61 75 74 6f 6d 61 74 | 69 63 61 6c 6c 79 20 61 | automat|ically a|
|000021a0| 76 6f 69 64 73 20 73 75 | 63 68 20 63 6f 6e 66 6c |voids su|ch confl|
|000021b0| 69 63 74 73 20 64 75 72 | 69 6e 67 20 74 68 65 20 |icts dur|ing the |
|000021c0| 6d 6f 64 65 6c 69 6e 67 | 20 70 72 6f 63 65 73 73 |modeling| process|
|000021d0| 20 61 6e 64 20 63 72 65 | 61 74 65 73 20 61 20 73 | and cre|ates a s|
|000021e0| 79 73 74 65 6d 20 77 68 | 69 63 68 20 77 65 20 77 |ystem wh|ich we w|
|000021f0| 69 6c 6c 20 63 61 6c 6c | 20 61 20 27 70 72 65 64 |ill call| a 'pred|
|00002200| 69 63 74 61 62 6c 65 20 | 73 79 73 74 65 6d 27 2e |ictable |system'.|
|00002210| 20 54 68 65 20 72 65 61 | 73 6f 6e 20 66 6f 72 20 | The rea|son for |
|00002220| 63 72 65 61 74 69 6e 67 | 20 64 79 6e 61 6d 69 63 |creating| dynamic|
|00002230| 20 73 79 73 74 65 6d 73 | 20 6f 66 20 65 71 75 61 | systems| of equa|
|00002240| 74 69 6f 6e 73 20 69 73 | 20 74 68 61 74 20 74 6f |tions is| that to|
|00002250| 20 70 72 65 64 69 63 74 | 20 61 6e 20 6f 75 74 70 | predict| an outp|
|00002260| 75 74 20 76 61 72 69 61 | 62 6c 65 20 69 74 20 69 |ut varia|ble it i|
|00002270| 73 20 6e 65 63 65 73 73 | 61 72 79 20 74 6f 20 68 |s necess|ary to h|
|00002280| 61 76 65 20 76 61 6c 75 | 65 73 20 6f 66 20 69 74 |ave valu|es of it|
|00002290| 73 20 69 6e 70 75 74 20 | 76 61 72 69 61 62 6c 65 |s input |variable|
|000022a0| 73 20 66 6f 72 20 74 68 | 65 20 66 6f 72 65 63 61 |s for th|e foreca|
|000022b0| 73 74 20 68 6f 72 69 7a | 6f 6e 20 61 76 61 69 6c |st horiz|on avail|
|000022c0| 61 62 6c 65 20 61 6c 73 | 6f 2e 20 54 68 65 73 65 |able als|o. These|
|000022d0| 20 76 61 6c 75 65 73 20 | 63 61 6e 20 62 65 20 65 | values |can be e|
|000022e0| 73 74 69 6d 61 74 65 64 | 20 6f 72 20 61 73 73 75 |stimated| or assu|
|000022f0| 6d 65 64 20 76 61 6c 75 | 65 73 20 28 77 68 69 63 |med valu|es (whic|
|00002300| 68 20 77 6f 75 6c 64 20 | 6c 65 61 64 20 74 6f 20 |h would |lead to |
|00002310| 61 20 77 68 61 74 2d 69 | 66 20 70 72 65 64 69 63 |a what-i|f predic|
|00002320| 74 69 6f 6e 29 20 6f 72 | 20 63 61 6e 20 62 65 20 |tion) or| can be |
|00002330| 70 72 65 64 69 63 74 65 | 64 20 61 6c 73 6f 20 62 |predicte|d also b|
|00002340| 79 20 74 69 6d 65 20 73 | 65 72 69 65 73 20 6d 6f |y time s|eries mo|
|00002350| 64 65 6c 73 20 6f 72 2c | 20 61 73 20 77 65 20 73 |dels or,| as we s|
|00002360| 75 67 67 65 73 74 2c 20 | 62 79 20 61 20 73 79 73 |uggest, |by a sys|
|00002370| 74 65 6d 20 6f 66 20 65 | 71 75 61 74 69 6f 6e 73 |tem of e|quations|
|00002380| 20 28 6c 65 61 64 69 6e | 67 20 74 6f 20 61 20 73 | (leadin|g to a s|
|00002390| 74 61 74 75 73 2d 71 75 | 6f 20 70 72 65 64 69 63 |tatus-qu|o predic|
|000023a0| 74 69 6f 6e 29 2e 0d 53 | 74 61 74 69 63 20 73 79 |tion)..S|tatic sy|
|000023b0| 73 74 65 6d 73 20 6f 66 | 20 65 71 75 61 74 69 6f |stems of| equatio|
|000023c0| 6e 73 20 63 61 6e 20 62 | 65 20 75 73 65 64 20 66 |ns can b|e used f|
|000023d0| 6f 72 20 61 6e 61 6c 79 | 73 69 73 2c 20 63 6c 61 |or analy|sis, cla|
|000023e0| 73 73 69 66 69 63 61 74 | 69 6f 6e 20 6f 72 20 64 |ssificat|ion or d|
|000023f0| 69 61 67 6e 6f 73 69 73 | 20 74 61 73 6b 73 2e 0d |iagnosis| tasks..|
|00002400| 0d 51 3a 20 57 68 61 74 | 20 69 73 20 69 6e 66 6f |.Q: What| is info|
|00002410| 72 6d 61 74 69 6f 6e 20 | 62 61 73 69 73 3f 20 49 |rmation |basis? I|
|00002420| 73 20 69 74 20 73 69 6d | 70 6c 79 20 61 6e 6f 74 |s it sim|ply anot|
|00002430| 68 65 72 20 77 6f 72 64 | 20 66 6f 72 20 74 68 65 |her word| for the|
|00002440| 20 64 61 74 61 20 62 61 | 73 69 73 20 73 74 6f 72 | data ba|sis stor|
|00002450| 65 64 20 69 6e 20 74 68 | 65 20 74 61 62 6c 65 3f |ed in th|e table?|
|00002460| 0d 41 3a 20 46 6f 72 20 | 64 79 6e 61 6d 69 63 20 |.A: For |dynamic |
|00002470| 6d 6f 64 65 6c 73 20 74 | 68 65 20 69 6e 66 6f 72 |models t|he infor|
|00002480| 6d 61 74 69 6f 6e 20 62 | 61 73 69 73 20 69 73 20 |mation b|asis is |
|00002490| 61 20 73 75 70 65 72 73 | 65 74 20 6f 66 20 74 68 |a supers|et of th|
|000024a0| 65 20 64 61 74 61 20 62 | 61 73 69 73 20 6c 6f 63 |e data b|asis loc|
|000024b0| 61 74 65 64 20 69 6e 20 | 74 68 65 20 74 61 62 6c |ated in |the tabl|
|000024c0| 65 2e 20 46 6f 72 20 73 | 74 61 74 69 63 20 6d 6f |e. For s|tatic mo|
|000024d0| 64 65 6c 73 20 74 68 65 | 79 20 61 72 65 20 65 71 |dels the|y are eq|
|000024e0| 75 61 6c 2e 20 54 68 65 | 20 69 6e 66 6f 72 6d 61 |ual. The| informa|
|000024f0| 74 69 6f 6e 20 62 61 73 | 69 73 20 69 73 20 75 73 |tion bas|is is us|
|00002500| 65 64 20 61 73 20 74 68 | 65 20 64 61 74 61 20 73 |ed as th|e data s|
|00002510| 6f 75 72 63 65 20 66 6f | 72 20 6d 6f 64 65 6c 69 |ource fo|r modeli|
|00002520| 6e 67 20 61 6e 64 20 69 | 74 20 63 6f 6e 74 61 69 |ng and i|t contai|
|00002530| 6e 73 20 61 6c 6c 20 63 | 68 6f 73 65 6e 20 75 6e |ns all c|hosen un|
|00002540| 6c 61 67 67 65 64 20 61 | 6e 64 20 6c 61 67 67 65 |lagged a|nd lagge|
|00002550| 64 20 76 61 72 69 61 62 | 6c 65 73 2e 20 55 73 75 |d variab|les. Usu|
|00002560| 61 6c 6c 79 2c 20 69 6e | 20 6f 74 68 65 72 20 6d |ally, in| other m|
|00002570| 6f 64 65 6c 69 6e 67 20 | 70 72 6f 67 72 61 6d 73 |odeling |programs|
|00002580| 2c 20 74 68 65 20 69 6e | 66 6f 72 6d 61 74 69 6f |, the in|formatio|
|00002590| 6e 20 62 61 73 69 73 20 | 6d 75 73 74 20 65 78 69 |n basis |must exi|
|000025a0| 73 74 20 70 68 79 73 69 | 63 61 6c 6c 79 20 69 6e |st physi|cally in|
|000025b0| 20 61 20 74 61 62 6c 65 | 2c 20 74 68 61 74 20 69 | a table|, that i|
|000025c0| 73 20 65 61 63 68 20 73 | 65 6c 65 63 74 65 64 20 |s each s|elected |
|000025d0| 28 75 6e 6c 61 67 67 65 | 64 20 6f 72 20 6c 61 67 |(unlagge|d or lag|
|000025e0| 67 65 64 29 20 69 6e 70 | 75 74 20 76 61 72 69 61 |ged) inp|ut varia|
|000025f0| 62 6c 65 20 6d 75 73 74 | 20 62 65 20 73 74 6f 72 |ble must| be stor|
|00002600| 65 64 20 69 6e 20 61 20 | 73 65 70 61 72 61 74 65 |ed in a |separate|
|00002610| 20 74 61 62 6c 65 20 63 | 6f 6c 75 6d 6e 2e 20 48 | table c|olumn. H|
|00002620| 65 72 65 20 69 6e 66 6f | 72 6d 61 74 69 6f 6e 20 |ere info|rmation |
|00002630| 62 61 73 69 73 20 61 6e | 64 20 64 61 74 61 20 62 |basis an|d data b|
|00002640| 61 73 69 73 20 6d 75 73 | 74 20 62 65 20 65 71 75 |asis mus|t be equ|
|00002650| 61 6c 2e 20 4c 61 67 67 | 65 64 20 69 6e 70 75 74 |al. Lagg|ed input|
|00002660| 20 76 61 72 69 61 62 6c | 65 73 20 61 73 20 64 65 | variabl|es as de|
|00002670| 72 69 76 61 74 69 76 65 | 73 20 6f 66 20 74 68 65 |rivative|s of the|
|00002680| 20 75 6e 6c 61 67 67 65 | 64 20 6f 6e 65 73 20 6d | unlagge|d ones m|
|00002690| 75 73 74 20 62 65 20 63 | 6f 6e 73 74 72 75 63 74 |ust be c|onstruct|
|000026a0| 65 64 20 6d 61 6e 75 61 | 6c 6c 79 20 62 79 20 63 |ed manua|lly by c|
|000026b0| 6f 70 79 69 6e 67 20 61 | 20 63 6f 6c 75 6d 6e 20 |opying a| column |
|000026c0| 61 6e 64 20 74 68 65 6e | 20 73 68 69 66 74 69 6e |and then| shiftin|
|000026d0| 67 20 69 74 20 6f 6e 65 | 20 6f 72 20 6d 6f 72 65 |g it one| or more|
|000026e0| 20 72 6f 77 73 20 64 6f | 77 6e 2e 20 4f 6e 63 65 | rows do|wn. Once|
|000026f0| 20 74 68 69 73 20 69 73 | 20 64 6f 6e 65 20 66 6f | this is| done fo|
|00002700| 72 20 65 61 63 68 20 6c | 61 67 67 65 64 20 76 61 |r each l|agged va|
|00002710| 72 69 61 62 6c 65 20 74 | 68 65 20 77 68 6f 6c 65 |riable t|he whole|
|00002720| 20 74 61 62 6c 65 20 6d | 75 73 74 20 62 65 20 73 | table m|ust be s|
|00002730| 68 69 66 74 65 64 20 61 | 67 61 69 6e 2c 20 75 70 |hifted a|gain, up|
|00002740| 77 61 72 64 73 20 6e 6f | 77 2e 20 49 6e 20 72 65 |wards no|w. In re|
|00002750| 73 75 6c 74 20 6f 66 20 | 74 68 69 73 20 74 69 6d |sult of |this tim|
|00002760| 65 20 61 6e 64 20 72 65 | 73 6f 75 72 63 65 20 69 |e and re|source i|
|00002770| 6e 74 65 6e 73 69 76 65 | 20 70 72 6f 63 65 73 73 |ntensive| process|
|00002780| 20 6f 6e 65 20 68 61 73 | 20 61 20 6c 61 72 67 65 | one has| a large|
|00002790| 2c 20 66 69 78 65 64 20 | 61 6e 64 20 68 69 67 68 |, fixed |and high|
|000027a0| 6c 79 20 72 65 64 75 6e | 64 61 6e 74 20 74 61 62 |ly redun|dant tab|
|000027b0| 6c 65 20 61 73 20 69 6e | 66 6f 72 6d 61 74 69 6f |le as in|formatio|
|000027c0| 6e 20 62 61 73 69 73 20 | 66 6f 72 20 6a 75 73 74 |n basis |for just|
|000027d0| 20 74 68 69 73 20 73 70 | 65 63 69 66 69 63 20 6d | this sp|ecific m|
|000027e0| 6f 64 65 6c 69 6e 67 20 | 70 72 6f 62 6c 65 6d 2e |odeling |problem.|
|000027f0| 20 0d 0d 46 6f 72 20 65 | 78 61 6d 70 6c 65 2c 20 | ..For e|xample, |
|00002800| 69 66 20 79 6f 75 20 77 | 61 6e 74 20 74 6f 20 63 |if you w|ant to c|
|00002810| 72 65 61 74 65 20 61 20 | 6d 6f 64 65 6c 20 6f 66 |reate a |model of|
|00002820| 20 35 20 69 6e 70 75 74 | 20 76 61 72 69 61 62 6c | 5 input| variabl|
|00002830| 65 73 20 61 6e 64 20 74 | 68 65 69 72 20 6c 61 67 |es and t|heir lag|
|00002840| 67 65 64 20 73 61 6d 70 | 6c 65 73 20 6f 66 20 75 |ged samp|les of u|
|00002850| 70 20 74 6f 20 35 20 74 | 68 65 20 69 6e 66 6f 72 |p to 5 t|he infor|
|00002860| 6d 61 74 69 6f 6e 20 62 | 61 73 69 73 20 77 69 6c |mation b|asis wil|
|00002870| 6c 20 63 6f 6e 73 69 73 | 74 20 6f 66 20 33 30 20 |l consis|t of 30 |
|00002880| 69 6e 70 75 74 20 76 61 | 72 69 61 62 6c 65 73 20 |input va|riables |
|00002890| 61 6c 6c 20 74 6f 67 65 | 74 68 65 72 20 28 74 68 |all toge|ther (th|
|000028a0| 65 20 35 20 75 6e 6c 61 | 67 67 65 64 20 61 6e 64 |e 5 unla|gged and|
|000028b0| 20 74 68 65 69 72 20 32 | 35 20 6c 61 67 67 65 64 | their 2|5 lagged|
|000028c0| 20 73 61 6d 70 6c 65 73 | 29 2e 20 59 6f 75 20 77 | samples|). You w|
|000028d0| 6f 75 6c 64 20 68 61 76 | 65 20 74 6f 20 65 78 74 |ould hav|e to ext|
|000028e0| 65 6e 64 20 74 68 65 20 | 74 61 62 6c 65 20 63 6f |end the |table co|
|000028f0| 6c 75 6d 6e 20 62 79 20 | 63 6f 6c 75 6d 6e 20 32 |lumn by |column 2|
|00002900| 35 20 74 69 6d 65 73 20 | 74 6f 20 67 65 74 20 74 |5 times |to get t|
|00002910| 68 65 20 64 65 73 69 72 | 65 64 20 69 6e 66 6f 72 |he desir|ed infor|
|00002920| 6d 61 74 69 6f 6e 20 62 | 61 73 69 73 20 6f 66 20 |mation b|asis of |
|00002930| 33 30 20 63 6f 6c 75 6d | 6e 73 2e 20 49 66 20 79 |30 colum|ns. If y|
|00002940| 6f 75 20 64 65 63 69 64 | 65 20 74 6f 20 63 6f 6e |ou decid|e to con|
|00002950| 73 69 64 65 72 20 61 6e | 20 61 64 64 69 74 69 6f |sider an| additio|
|00002960| 6e 61 6c 20 76 61 72 69 | 61 62 6c 65 20 6f 72 20 |nal vari|able or |
|00002970| 74 6f 20 63 72 65 61 74 | 65 20 61 20 6d 6f 64 65 |to creat|e a mode|
|00002980| 6c 20 66 6f 72 20 61 6e | 6f 74 68 65 72 20 6f 75 |l for an|other ou|
|00002990| 74 70 75 74 20 76 61 72 | 69 61 62 6c 65 20 79 6f |tput var|iable yo|
|000029a0| 75 20 68 61 76 65 20 74 | 6f 20 73 74 6f 72 65 20 |u have t|o store |
|000029b0| 74 68 65 20 63 6f 6d 70 | 6c 65 74 65 20 66 69 72 |the comp|lete fir|
|000029c0| 73 74 20 74 61 62 6c 65 | 20 61 6e 64 20 74 68 65 |st table| and the|
|000029d0| 6e 20 72 65 6f 72 67 61 | 6e 69 7a 65 20 69 74 20 |n reorga|nize it |
|000029e0| 61 63 63 6f 72 64 69 6e | 67 6c 79 20 74 6f 20 74 |accordin|gly to t|
|000029f0| 68 65 20 6e 65 77 20 6d | 6f 64 65 6c 69 6e 67 20 |he new m|odeling |
|00002a00| 70 72 6f 62 6c 65 6d 20 | 69 6e 20 74 68 65 20 77 |problem |in the w|
|00002a10| 61 79 20 64 65 73 63 72 | 69 62 65 64 20 61 62 6f |ay descr|ibed abo|
|00002a20| 76 65 2e 20 45 61 63 68 | 20 6d 6f 64 65 6c 20 6e |ve. Each| model n|
|00002a30| 65 65 64 73 20 74 6f 20 | 68 61 76 65 20 69 74 73 |eeds to |have its|
|00002a40| 20 6f 77 6e 2c 20 70 68 | 79 73 69 63 61 6c 20 69 | own, ph|ysical i|
|00002a50| 6e 66 6f 72 6d 61 74 69 | 6f 6e 20 62 61 73 69 73 |nformati|on basis|
|00002a60| 2e 20 54 68 69 73 20 69 | 73 20 74 68 65 20 77 61 |. This i|s the wa|
|00002a70| 79 20 69 74 20 77 6f 72 | 6b 73 20 6f 6e 20 44 4f |y it wor|ks on DO|
|00002a80| 53 20 6f 72 20 57 69 6e | 64 6f 77 73 20 6d 61 63 |S or Win|dows mac|
|00002a90| 68 69 6e 65 73 2e 0d 0d | 57 65 20 68 61 76 65 20 |hines...|We have |
|00002aa0| 61 73 6b 20 6f 75 72 73 | 65 6c 76 65 73 20 77 68 |ask ours|elves wh|
|00002ab0| 79 20 74 68 65 20 69 6e | 66 6f 72 6d 61 74 69 6f |y the in|formatio|
|00002ac0| 6e 20 62 61 73 69 73 20 | 28 61 6e 64 20 66 69 6e |n basis |(and fin|
|00002ad0| 61 6c 6c 79 20 74 68 65 | 20 75 73 65 72 29 20 68 |ally the| user) h|
|00002ae0| 61 73 20 74 6f 20 6d 65 | 65 74 20 74 68 65 20 70 |as to me|et the p|
|00002af0| 72 6f 67 72 61 6d 27 73 | 20 6e 65 65 64 73 20 61 |rogram's| needs a|
|00002b00| 6e 64 20 6e 6f 74 2c 20 | 63 6f 6e 76 65 72 73 65 |nd not, |converse|
|00002b10| 6c 79 2c 20 74 68 65 20 | 70 72 6f 67 72 61 6d 20 |ly, the |program |
|00002b20| 68 61 73 20 74 6f 20 6d | 65 65 74 20 74 68 65 20 |has to m|eet the |
|00002b30| 74 61 62 6c 65 27 73 20 | 73 74 61 74 65 3f 20 57 |table's |state? W|
|00002b40| 65 20 66 69 6e 64 20 74 | 68 61 74 20 74 68 65 20 |e find t|hat the |
|00002b50| 75 73 65 72 20 6d 75 73 | 74 20 62 65 20 61 62 6c |user mus|t be abl|
|00002b60| 65 20 61 74 20 61 6e 79 | 20 74 69 6d 65 20 61 6e |e at any| time an|
|00002b70| 64 20 66 6f 72 20 61 6e | 79 20 63 6f 6e 74 69 6e |d for an|y contin|
|00002b80| 75 6f 75 73 20 6f 72 20 | 64 69 73 63 6f 6e 74 69 |uous or |disconti|
|00002b90| 6e 75 6f 75 73 20 63 6f | 72 65 20 64 61 74 61 20 |nuous co|re data |
|00002ba0| 62 61 73 69 73 20 74 6f | 20 63 72 65 61 74 65 20 |basis to| create |
|00002bb0| 6d 6f 64 65 6c 73 20 6f | 75 74 20 6f 66 20 61 6e |models o|ut of an|
|00002bc0| 79 20 70 6f 73 73 69 62 | 6c 65 20 69 6e 70 75 74 |y possib|le input|
|00002bd0| 20 76 61 72 69 61 62 6c | 65 20 63 6f 6d 62 69 6e | variabl|e combin|
|00002be0| 61 74 69 6f 6e 20 66 72 | 6f 6d 20 74 68 65 20 64 |ation fr|om the d|
|00002bf0| 61 74 61 20 62 61 73 69 | 73 20 61 73 20 69 73 2e |ata basi|s as is.|
|00002c00| 20 54 68 69 73 20 6d 65 | 61 6e 73 2c 20 4b 6e 6f | This me|ans, Kno|
|00002c10| 77 6c 65 64 67 65 4d 69 | 6e 65 72 20 63 72 65 61 |wledgeMi|ner crea|
|00002c20| 74 65 73 20 66 6f 72 20 | 79 6f 75 20 61 6e 20 69 |tes for |you an i|
|00002c30| 6e 66 6f 72 6d 61 74 69 | 6f 6e 20 62 61 73 69 73 |nformati|on basis|
|00002c40| 20 61 75 74 6f 6d 61 74 | 69 63 61 6c 6c 79 20 61 | automat|ically a|
|00002c50| 6e 64 20 76 69 72 74 75 | 61 6c 6c 79 20 75 73 69 |nd virtu|ally usi|
|00002c60| 6e 67 20 74 68 65 20 64 | 61 74 61 20 70 68 79 73 |ng the d|ata phys|
|00002c70| 69 63 61 6c 6c 79 20 73 | 74 6f 72 65 64 20 69 6e |ically s|tored in|
|00002c80| 20 74 68 65 20 74 61 62 | 6c 65 2e 20 49 6e 20 74 | the tab|le. In t|
|00002c90| 68 69 73 20 77 61 79 20 | 69 74 20 69 73 20 70 6f |his way |it is po|
|00002ca0| 73 73 69 62 6c 65 20 74 | 6f 20 63 72 65 61 74 65 |ssible t|o create|
|00002cb0| 20 74 68 65 6f 72 65 74 | 69 63 61 6c 6c 79 20 61 | theoret|ically a|
|00002cc0| 6e 20 69 6e 66 69 6e 69 | 74 65 20 6e 75 6d 62 65 |n infini|te numbe|
|00002cd0| 72 20 6f 66 20 6d 6f 64 | 65 6c 73 20 75 73 69 6e |r of mod|els usin|
|00002ce0| 67 20 6f 6e 65 20 61 6e | 64 20 74 68 65 20 73 61 |g one an|d the sa|
|00002cf0| 6d 65 20 64 61 74 61 20 | 62 61 73 69 73 20 77 69 |me data |basis wi|
|00002d00| 74 68 6f 75 74 20 61 6e | 79 20 63 68 61 6e 67 65 |thout an|y change|
|00002d10| 2e 20 59 6f 75 20 6f 6e | 6c 79 20 68 61 76 65 20 |. You on|ly have |
|00002d20| 74 6f 20 73 65 6c 65 63 | 74 20 74 68 65 20 76 61 |to selec|t the va|
|00002d30| 72 69 61 62 6c 65 73 20 | 69 6e 20 74 68 65 20 74 |riables |in the t|
|00002d40| 61 62 6c 65 20 79 6f 75 | 20 77 61 6e 74 20 74 6f |able you| want to|
|00002d50| 20 75 73 65 20 66 6f 72 | 20 6d 6f 64 65 6c 69 6e | use for| modelin|
|00002d60| 67 20 69 6e 20 61 20 64 | 65 66 69 6e 65 64 20 77 |g in a d|efined w|
|00002d70| 61 79 2e 20 55 73 69 6e | 67 20 74 68 65 20 65 78 |ay. Usin|g the ex|
|00002d80| 61 6d 70 6c 65 20 61 62 | 6f 76 65 2c 20 79 6f 75 |ample ab|ove, you|
|00002d90| 20 77 6f 75 6c 64 20 68 | 61 76 65 20 74 6f 20 73 | would h|ave to s|
|00002da0| 65 6c 65 63 74 20 74 68 | 65 20 66 69 72 73 74 20 |elect th|e first |
|00002db0| 36 20 72 6f 77 73 20 6f | 66 20 65 61 63 68 20 63 |6 rows o|f each c|
|00002dc0| 6f 6c 75 6d 6e 20 77 68 | 69 63 68 20 68 61 73 20 |olumn wh|ich has |
|00002dd0| 73 74 6f 72 65 64 20 74 | 68 65 20 64 61 74 61 20 |stored t|he data |
|00002de0| 6f 66 20 6f 6e 65 20 6f | 66 20 74 68 65 20 35 20 |of one o|f the 5 |
|00002df0| 76 61 72 69 61 62 6c 65 | 73 20 79 6f 75 20 77 61 |variable|s you wa|
|00002e00| 6e 74 20 74 6f 20 75 73 | 65 20 28 65 2e 67 2e 20 |nt to us|e (e.g. |
|00002e10| 63 6f 6c 75 6d 6e 20 31 | 2c 20 34 2c 20 31 32 2c |column 1|, 4, 12,|
|00002e20| 20 31 33 20 61 6e 64 20 | 32 35 29 2e 20 59 6f 75 | 13 and |25). You|
|00002e30| 20 64 6f 6e 27 74 20 68 | 61 76 65 20 74 6f 20 63 | don't h|ave to c|
|00002e40| 61 72 65 20 61 62 6f 75 | 74 20 74 68 65 20 72 65 |are abou|t the re|
|00002e50| 73 74 2e 20 49 66 20 79 | 6f 75 20 64 65 63 69 64 |st. If y|ou decid|
|00002e60| 65 20 74 6f 20 63 72 65 | 61 74 65 20 61 6e 20 61 |e to cre|ate an a|
|00002e70| 6c 74 65 72 6e 61 74 69 | 76 65 20 6d 6f 64 65 6c |lternati|ve model|
|00002e80| 20 75 73 69 6e 67 20 61 | 20 36 74 68 20 76 61 72 | using a| 6th var|
|00002e90| 69 61 62 6c 65 20 61 6c | 73 6f 20 79 6f 75 20 73 |iable al|so you s|
|00002ea0| 69 6d 70 6c 79 20 68 61 | 76 65 20 74 6f 20 73 65 |imply ha|ve to se|
|00002eb0| 6c 65 63 74 20 36 20 61 | 64 64 69 74 69 6f 6e 61 |lect 6 a|dditiona|
|00002ec0| 6c 20 63 65 6c 6c 73 20 | 69 6e 20 74 68 65 20 74 |l cells |in the t|
|00002ed0| 61 62 6c 65 20 61 6e 64 | 20 74 68 65 6e 20 6c 65 |able and| then le|
|00002ee0| 74 20 74 68 65 20 63 6f | 6d 70 75 74 65 72 20 64 |t the co|mputer d|
|00002ef0| 6f 20 74 68 65 20 72 65 | 73 74 2e 20 41 6e 64 20 |o the re|st. And |
|00002f00| 62 6f 74 68 20 6d 6f 64 | 65 6c 73 20 77 69 6c 6c |both mod|els will|
|00002f10| 20 77 6f 72 6b 20 6f 6e | 20 74 68 65 20 73 61 6d | work on| the sam|
|00002f20| 65 20 6f 6c 64 20 64 61 | 74 61 20 62 61 73 69 73 |e old da|ta basis|
|00002f30| 20 61 6c 74 68 6f 75 67 | 68 20 74 68 65 79 20 75 | althoug|h they u|
|00002f40| 73 65 20 61 20 64 69 66 | 66 65 72 65 6e 74 20 69 |se a dif|ferent i|
|00002f50| 6e 66 6f 72 6d 61 74 69 | 6f 6e 20 62 61 73 69 73 |nformati|on basis|
|00002f60| 2e 0d 0d 43 68 6f 6f 73 | 69 6e 67 20 74 68 65 20 |...Choos|ing the |
|00002f70| 69 6e 66 6f 72 6d 61 74 | 69 6f 6e 20 62 61 73 69 |informat|ion basi|
|00002f80| 73 20 69 73 20 74 68 65 | 20 75 73 65 72 73 20 6d |s is the| users m|
|00002f90| 6f 73 74 20 69 6d 70 6f | 72 74 61 6e 74 20 74 61 |ost impo|rtant ta|
|00002fa0| 73 6b 20 62 65 66 6f 72 | 65 20 63 72 65 61 74 69 |sk befor|e creati|
|00002fb0| 6e 67 20 61 20 6d 6f 64 | 65 6c 2e 20 47 65 6e 65 |ng a mod|el. Gene|
|00002fc0| 72 61 6c 6c 79 2c 20 79 | 6f 75 20 68 61 76 65 20 |rally, y|ou have |
|00002fd0| 74 6f 20 61 6e 73 77 65 | 72 20 74 68 65 20 71 75 |to answe|r the qu|
|00002fe0| 65 73 74 69 6f 6e 20 27 | 42 79 20 77 68 69 63 68 |estion '|By which|
|00002ff0| 20 69 6e 70 75 74 20 76 | 61 72 69 61 62 6c 65 73 | input v|ariables|
|00003000| 20 63 6f 75 6c 64 20 6d | 79 20 6f 75 74 70 75 74 | could m|y output|
|00003010| 20 76 61 72 69 61 62 6c | 65 20 62 65 20 61 66 66 | variabl|e be aff|
|00003020| 65 63 74 65 64 3f 27 2e | 20 43 68 6f 6f 73 69 6e |ected?'.| Choosin|
|00003030| 67 20 74 68 65 20 69 6e | 66 6f 72 6d 61 74 69 6f |g the in|formatio|
|00003040| 6e 20 62 61 73 69 73 20 | 69 73 20 6f 6e 6c 79 20 |n basis |is only |
|00003050| 61 6e 20 61 73 73 75 6d | 70 74 69 6f 6e 20 74 68 |an assum|ption th|
|00003060| 61 74 20 61 6e 79 20 64 | 65 70 65 6e 64 65 6e 63 |at any d|ependenc|
|00003070| 65 20 63 6f 75 6c 64 20 | 62 65 20 65 78 69 73 74 |e could |be exist|
|00003080| 2e 20 49 74 20 64 6f 6e | 27 74 20 70 72 65 64 65 |. It don|'t prede|
|00003090| 66 69 6e 65 73 20 74 68 | 65 20 66 69 6e 61 6c 20 |fines th|e final |
|000030a0| 6d 6f 64 65 6c 20 73 74 | 72 75 63 74 75 72 65 2e |model st|ructure.|
|000030b0| 20 49 74 20 64 65 66 69 | 6e 65 73 20 6f 6e 6c 79 | It defi|nes only|
|000030c0| 20 74 68 69 73 20 73 65 | 74 20 6f 66 20 76 61 72 | this se|t of var|
|000030d0| 69 61 62 6c 65 73 20 66 | 72 6f 6d 20 77 68 69 63 |iables f|rom whic|
|000030e0| 68 20 61 20 73 75 62 73 | 65 74 20 6f 66 20 72 65 |h a subs|et of re|
|000030f0| 61 6c 6c 79 20 72 65 6c | 65 76 61 6e 74 20 76 61 |ally rel|evant va|
|00003100| 72 69 61 62 6c 65 73 20 | 77 69 6c 6c 20 62 65 20 |riables |will be |
|00003110| 75 73 65 64 20 69 6e 20 | 74 68 65 20 66 69 6e 61 |used in |the fina|
|00003120| 6c 20 6d 6f 64 65 6c 20 | 28 6b 6e 6f 77 6c 65 64 |l model |(knowled|
|00003130| 67 65 20 65 78 74 72 61 | 63 74 69 6f 6e 29 2e 0d |ge extra|ction)..|
|00003140| 0d 51 3a 20 49 20 77 61 | 73 20 69 6d 70 72 65 73 |.Q: I wa|s impres|
|00003150| 73 65 64 20 62 79 20 79 | 6f 75 72 20 64 65 73 63 |sed by y|our desc|
|00003160| 72 69 70 74 69 6f 6e 20 | 6f 66 20 4b 6e 6f 77 6c |ription |of Knowl|
|00003170| 65 64 67 65 4d 69 6e 65 | 72 2e 20 41 63 74 75 61 |edgeMine|r. Actua|
|00003180| 6c 6c 79 20 69 74 20 73 | 65 65 6d 73 20 74 6f 20 |lly it s|eems to |
|00003190| 66 69 6e 64 20 6f 70 74 | 69 6d 61 6c 20 6e 65 75 |find opt|imal neu|
|000031a0| 72 61 6c 20 6e 65 74 73 | 20 2d 20 61 20 74 61 73 |ral nets| - a tas|
|000031b0| 6b 20 49 20 74 68 6f 75 | 67 68 74 20 77 61 73 20 |k I thou|ght was |
|000031c0| 6d 6f 72 65 20 6f 72 20 | 6c 65 73 73 20 69 6d 70 |more or |less imp|
|000031d0| 6f 73 73 69 62 6c 65 2e | 20 20 49 20 77 61 6e 74 |ossible.| I want|
|000031e0| 20 74 6f 20 75 73 65 20 | 69 74 20 69 6e 20 6c 69 | to use |it in li|
|000031f0| 6e 67 75 69 73 74 69 63 | 73 20 2d 20 49 20 70 6c |nguistic|s - I pl|
|00003200| 61 6e 6e 65 64 20 74 6f | 20 74 61 6b 65 20 61 20 |anned to| take a |
|00003210| 76 65 63 74 6f 72 20 72 | 65 70 72 65 73 65 6e 74 |vector r|epresent|
|00003220| 61 74 69 6f 6e 20 6f 66 | 20 61 20 77 6f 72 64 20 |ation of| a word |
|00003230| 61 6e 64 20 6d 61 70 20 | 74 68 61 74 20 69 6e 74 |and map |that int|
|00003240| 6f 20 61 20 72 65 70 72 | 65 73 65 6e 74 61 74 69 |o a repr|esentati|
|00003250| 6f 6e 20 6f 66 20 61 6e | 6f 74 68 65 72 20 66 6f |on of an|other fo|
|00003260| 72 6d 20 6f 66 20 74 68 | 61 74 20 77 6f 72 64 2e |rm of th|at word.|
|00003270| 20 49 20 6a 75 73 74 20 | 68 61 76 65 20 6f 6e 65 | I just |have one|
|00003280| 20 71 75 65 73 74 69 6f | 6e 2e 20 0d 2a 20 44 6f | questio|n. .* Do|
|00003290| 65 73 20 4b 4d 20 61 73 | 73 75 6d 65 20 74 68 61 |es KM as|sume tha|
|000032a0| 74 20 74 68 65 72 65 20 | 69 73 20 61 20 6f 6e 65 |t there |is a one|
|000032b0| 2d 74 6f 2d 6f 6e 65 20 | 6f 72 20 6d 61 6e 79 2d |-to-one |or many-|
|000032c0| 74 6f 2d 6f 6e 65 20 72 | 65 6c 61 74 69 6f 6e 20 |to-one r|elation |
|000032d0| 69 6e 20 0d 74 68 65 20 | 64 61 74 61 20 6f 72 20 |in .the |data or |
|000032e0| 63 61 6e 20 69 74 20 73 | 75 70 70 6f 72 74 20 6f |can it s|upport o|
|000032f0| 6e 65 2d 74 6f 2d 6d 61 | 6e 79 20 6d 61 70 70 69 |ne-to-ma|ny mappi|
|00003300| 6e 67 73 20 61 73 20 77 | 65 6c 6c 3f 0d 41 3a 20 |ngs as w|ell?.A: |
|00003310| 49 20 77 69 6c 6c 20 74 | 72 79 20 74 6f 20 61 6e |I will t|ry to an|
|00003320| 73 77 65 72 20 79 6f 75 | 72 20 71 75 65 73 74 69 |swer you|r questi|
|00003330| 6f 6e 20 62 75 74 20 49 | 27 6d 20 6e 6f 74 20 73 |on but I|'m not s|
|00003340| 75 72 65 20 69 66 20 49 | 20 75 6e 64 65 72 73 74 |ure if I| underst|
|00003350| 61 6e 64 20 63 6f 72 72 | 65 63 74 6c 79 2e 0d 46 |and corr|ectly..F|
|00003360| 6f 72 20 65 61 63 68 20 | 76 61 72 69 61 62 6c 65 |or each |variable|
|00003370| 20 28 63 6f 6c 75 6d 6e | 29 20 79 6f 75 20 63 61 | (column|) you ca|
|00003380| 6e 20 63 72 65 61 74 65 | 20 61 6e 64 20 73 74 6f |n create| and sto|
|00003390| 72 65 20 0d 20 20 20 20 | 20 20 20 2d 20 61 20 74 |re . | - a t|
|000033a0| 69 6d 65 20 73 65 72 69 | 65 73 20 6d 6f 64 65 6c |ime seri|es model|
|000033b0| 20 28 61 75 74 6f 2d 72 | 65 67 72 65 73 73 69 76 | (auto-r|egressiv|
|000033c0| 65 20 6d 6f 64 65 6c 29 | 20 69 6e 20 74 68 65 20 |e model)| in the |
|000033d0| 66 6f 72 6d 0d 20 20 20 | 20 20 20 20 20 20 20 20 |form. | |
|000033e0| 20 20 20 20 20 20 20 59 | 74 20 3d 20 66 28 59 74 | Y|t = f(Yt|
|000033f0| 2d 31 2c 20 59 74 2d 32 | 2c 20 2e 2e 2e 2c 20 59 |-1, Yt-2|, ..., Y|
|00003400| 74 2d 6e 29 20 61 74 20 | 61 20 67 69 76 65 6e 20 |t-n) at |a given |
|00003410| 74 69 6d 65 20 74 3b 0d | 20 20 20 20 20 20 20 2d |time t;.| -|
|00003420| 20 61 20 6d 75 6c 74 69 | 2d 69 6e 70 75 74 2f 73 | a multi|-input/s|
|00003430| 69 6e 67 6c 65 20 6f 75 | 74 70 75 74 20 6d 6f 64 |ingle ou|tput mod|
|00003440| 65 6c 20 69 6e 20 74 68 | 65 20 66 6f 72 6d 0d 20 |el in th|e form. |
|00003450| 20 20 20 20 20 20 20 20 | 20 20 20 20 20 20 20 20 | | |
|00003460| 20 20 59 20 3d 20 66 28 | 58 31 2c 20 58 32 2c 20 | Y = f(|X1, X2, |
|00003470| 2e 2e 2e 2c 20 58 6d 29 | 0d 77 68 65 72 65 62 79 |..., Xm)|.whereby|
|00003480| 20 59 20 69 73 20 74 68 | 65 20 6f 75 74 70 75 74 | Y is th|e output|
|00003490| 20 76 61 72 69 61 62 6c | 65 20 76 65 63 74 6f 72 | variabl|e vector|
|000034a0| 2c 20 58 6d 20 74 68 65 | 20 69 6e 70 75 74 20 76 |, Xm the| input v|
|000034b0| 61 72 69 61 62 6c 65 20 | 76 65 63 74 6f 72 73 20 |ariable |vectors |
|000034c0| 61 6e 64 20 66 20 74 68 | 65 20 75 6e 6b 6e 6f 77 |and f th|e unknow|
|000034d0| 6e 20 72 65 6c 61 74 69 | 6f 6e 20 62 65 74 77 65 |n relati|on betwe|
|000034e0| 65 6e 20 74 68 65 73 65 | 20 76 61 72 69 61 62 6c |en these| variabl|
|000034f0| 65 73 2e 20 54 68 65 20 | 74 61 73 6b 20 6f 66 20 |es. The |task of |
|00003500| 4b 6e 6f 77 6c 65 64 67 | 65 4d 69 6e 65 72 20 69 |Knowledg|eMiner i|
|00003510| 73 20 74 6f 20 63 72 65 | 61 74 65 20 61 20 6c 61 |s to cre|ate a la|
|00003520| 72 67 65 20 6e 75 6d 62 | 65 72 20 6f 66 20 61 6c |rge numb|er of al|
|00003530| 74 65 72 6e 61 74 69 76 | 65 20 72 65 6c 61 74 69 |ternativ|e relati|
|00003540| 6f 6e 73 20 66 20 6f 66 | 20 69 6e 63 72 65 61 73 |ons f of| increas|
|00003550| 69 6e 67 20 63 6f 6d 70 | 6c 65 78 69 74 79 20 61 |ing comp|lexity a|
|00003560| 6e 64 20 74 6f 20 73 65 | 6c 65 63 74 20 61 20 62 |nd to se|lect a b|
|00003570| 65 73 74 2c 20 76 61 6c | 69 64 20 72 65 6c 61 74 |est, val|id relat|
|00003580| 69 6f 6e 20 66 20 61 63 | 63 6f 72 64 69 6e 67 20 |ion f ac|cording |
|00003590| 74 6f 20 61 20 67 69 76 | 65 6e 20 63 72 69 74 65 |to a giv|en crite|
|000035a0| 72 69 6f 6e 2e 20 4f 72 | 2c 20 6d 6f 72 65 20 63 |rion. Or|, more c|
|000035b0| 6f 72 72 65 63 74 6c 79 | 2c 20 74 68 65 20 74 68 |orrectly|, the th|
|000035c0| 72 65 65 20 62 65 73 74 | 20 6d 6f 64 65 6c 73 20 |ree best| models |
|000035d0| 61 72 65 20 73 65 6c 65 | 63 74 65 64 20 61 6e 64 |are sele|cted and|
|000035e0| 20 73 74 6f 72 65 64 20 | 69 6e 20 74 68 65 20 6d | stored |in the m|
|000035f0| 6f 64 65 6c 20 62 61 73 | 65 2e 20 0d 54 68 65 20 |odel bas|e. .The |
|00003600| 58 6d 27 73 20 63 61 6e | 20 62 65 20 6f 62 73 65 |Xm's can| be obse|
|00003610| 72 76 61 74 69 6f 6e 61 | 6c 20 64 61 74 61 2c 20 |rvationa|l data, |
|00003620| 74 69 6d 65 20 6c 61 67 | 67 65 64 20 73 61 6d 70 |time lag|ged samp|
|00003630| 6c 65 73 20 6f 66 20 6f | 62 73 65 72 76 65 64 20 |les of o|bserved |
|00003640| 76 61 72 69 61 62 6c 65 | 73 20 6f 72 20 66 75 6e |variable|s or fun|
|00003650| 63 74 69 6f 6e 73 20 6f | 66 20 58 6d 20 28 73 69 |ctions o|f Xm (si|
|00003660| 6e 28 78 29 2c 20 65 78 | 70 28 78 29 20 65 2e 67 |n(x), ex|p(x) e.g|
|00003670| 2e 29 2e 20 59 3d 66 28 | 58 31 2c 58 32 2c 2e 2e |.). Y=f(|X1,X2,..|
|00003680| 2e 58 6d 29 20 63 61 6e | 20 65 69 74 68 65 72 20 |.Xm) can| either |
|00003690| 62 65 20 61 20 73 74 61 | 74 69 63 20 6d 6f 64 65 |be a sta|tic mode|
|000036a0| 6c 20 6f 72 20 61 20 64 | 79 6e 61 6d 69 63 20 6d |l or a d|ynamic m|
|000036b0| 6f 64 65 6c 20 69 66 20 | 74 69 6d 65 20 6c 61 67 |odel if |time lag|
|000036c0| 67 65 64 20 73 61 6d 70 | 6c 65 73 20 61 72 65 20 |ged samp|les are |
|000036d0| 75 73 65 64 2e 20 54 68 | 65 6e 20 74 68 65 20 6d |used. Th|en the m|
|000036e0| 6f 64 65 6c 20 69 73 20 | 72 65 66 6c 65 63 74 69 |odel is |reflecti|
|000036f0| 6e 67 20 61 20 64 69 66 | 66 65 72 65 6e 63 65 20 |ng a dif|ference |
|00003700| 65 71 75 61 74 69 6f 6e | 2e 0d 0d 20 20 20 20 20 |equation|... |
|00003710| 20 20 20 2d 20 61 20 6d | 75 6c 74 69 2d 69 6e 70 | - a m|ulti-inp|
|00003720| 75 74 2f 73 69 6e 67 6c | 65 20 6f 75 74 70 75 74 |ut/singl|e output|
|00003730| 20 6d 6f 64 65 6c 20 61 | 73 20 70 61 72 74 20 6f | model a|s part o|
|00003740| 66 20 61 20 6d 75 6c 74 | 69 2d 69 6e 70 75 74 2f |f a mult|i-input/|
|00003750| 6d 75 6c 74 69 2d 6f 75 | 74 70 75 74 20 73 79 73 |multi-ou|tput sys|
|00003760| 74 65 6d 20 28 73 79 73 | 74 65 6d 20 6f 66 0d 20 |tem (sys|tem of. |
|00003770| 20 20 20 20 20 20 20 20 | 20 65 71 75 61 74 69 6f | | equatio|
|00003780| 6e 73 29 20 77 68 69 63 | 68 20 6d 61 79 20 68 61 |ns) whic|h may ha|
|00003790| 76 65 20 74 68 65 20 66 | 6f 72 6d 0d 20 20 20 20 |ve the f|orm. |
|000037a0| 20 20 20 20 20 20 20 20 | 20 20 20 20 20 20 20 58 | | X|
|000037b0| 31 3d 66 31 28 58 32 2c | 2e 2e 2e 58 6d 29 0d 20 |1=f1(X2,|...Xm). |
|000037c0| 20 20 20 20 20 20 20 20 | 20 20 20 20 20 20 20 20 | | |
|000037d0| 20 20 58 32 3d 66 32 28 | 58 31 2c 58 33 2e 2e 2e | X2=f2(|X1,X3...|
|000037e0| 58 6d 29 0d 20 20 20 20 | 20 20 20 20 20 20 20 20 |Xm). | |
|000037f0| 20 20 20 20 20 20 20 20 | 2e 2e 2e 2e 2e 0d 20 20 | |...... |
|00003800| 20 20 20 20 20 20 20 20 | 20 20 20 20 20 20 20 20 | | |
|00003810| 20 58 6d 3d 66 6d 28 58 | 31 2c 58 32 2c 2e 2e 2e | Xm=fm(X|1,X2,...|
|00003820| 58 6d 2d 31 29 0d 46 6f | 72 20 61 20 64 79 6e 61 |Xm-1).Fo|r a dyna|
|00003830| 6d 69 63 20 73 79 73 74 | 65 6d 20 77 65 20 6f 62 |mic syst|em we ob|
|00003840| 74 61 69 6e 20 61 6e 20 | 61 75 74 6f 6e 6f 6d 6f |tain an |autonomo|
|00003850| 75 73 20 73 79 73 74 65 | 6d 20 6f 66 20 6d 20 64 |us syste|m of m d|
|00003860| 69 66 66 65 72 65 6e 63 | 65 20 65 71 75 61 74 69 |ifferenc|e equati|
|00003870| 6f 6e 73 20 72 65 73 70 | 65 63 74 69 76 65 6c 79 |ons resp|ectively|
|00003880| 20 77 68 69 63 68 20 69 | 73 20 61 70 70 6c 69 63 | which i|s applic|
|00003890| 61 62 6c 65 20 66 6f 72 | 20 73 74 61 74 75 73 2d |able for| status-|
|000038a0| 71 75 6f 20 70 72 65 64 | 69 63 74 69 6f 6e 2e 20 |quo pred|iction. |
|000038b0| 53 75 63 68 20 61 20 73 | 79 73 74 65 6d 20 72 65 |Such a s|ystem re|
|000038c0| 66 6c 65 63 74 73 20 74 | 68 65 20 69 6e 74 65 72 |flects t|he inter|
|000038d0| 63 6f 6e 6e 65 63 74 69 | 6f 6e 73 20 62 65 74 77 |connecti|ons betw|
|000038e0| 65 65 6e 20 61 6c 6c 20 | 76 61 72 69 61 62 6c 65 |een all |variable|
|000038f0| 73 20 61 6e 64 20 69 73 | 20 76 69 73 69 62 6c 65 |s and is| visible|
|00003900| 20 74 68 72 6f 75 67 68 | 20 61 20 73 79 73 74 65 | through| a syste|
|00003910| 6d 20 67 72 61 70 68 2e | 0d 0d 51 3a 20 57 6f 75 |m graph.|..Q: Wou|
|00003920| 6c 64 20 79 6f 75 20 63 | 6f 6e 73 69 64 65 72 20 |ld you c|onsider |
|00003930| 79 6f 75 72 20 70 72 6f | 64 75 63 74 73 20 73 75 |your pro|ducts su|
|00003940| 69 74 61 62 6c 65 20 66 | 6f 72 20 66 69 6e 61 6e |itable f|or finan|
|00003950| 63 69 61 6c 20 61 6e 64 | 20 70 72 6f 64 75 63 74 |cial and| product|
|00003960| 2d 64 65 6d 61 6e 64 20 | 66 6f 72 65 63 61 73 74 |-demand |forecast|
|00003970| 69 6e 67 20 75 73 69 6e | 67 20 6e 75 6d 65 72 6f |ing usin|g numero|
|00003980| 75 73 20 76 61 72 69 61 | 62 6c 65 20 69 6e 70 75 |us varia|ble inpu|
|00003990| 74 73 3f 0d 41 3a 20 59 | 65 73 2c 20 74 68 69 73 |ts?.A: Y|es, this|
|000039a0| 20 69 73 20 6f 6e 65 20 | 6f 66 20 74 68 65 20 70 | is one |of the p|
|000039b0| 72 69 6d 61 72 79 20 61 | 70 70 6c 69 63 61 74 69 |rimary a|pplicati|
|000039c0| 6f 6e 20 66 69 65 6c 64 | 73 20 6f 66 20 4b 6e 6f |on field|s of Kno|
|000039d0| 77 6c 65 64 67 65 4d 69 | 6e 65 72 2e 20 49 6e 20 |wledgeMi|ner. In |
|000039e0| 63 6f 6e 74 72 61 73 74 | 20 74 6f 20 73 74 61 74 |contrast| to stat|
|000039f0| 69 73 74 69 63 73 20 6f | 72 20 4e 4e 27 73 20 79 |istics o|r NN's y|
|00003a00| 6f 75 20 63 61 6e 20 75 | 73 65 20 6d 6f 72 65 20 |ou can u|se more |
|00003a10| 76 61 72 69 61 62 6c 65 | 73 20 74 68 61 6e 20 73 |variable|s than s|
|00003a20| 61 6d 70 6c 65 73 20 61 | 76 61 69 6c 61 62 6c 65 |amples a|vailable|
|00003a30| 2e 20 46 6f 72 20 65 78 | 61 6d 70 6c 65 2c 20 79 |. For ex|ample, y|
|00003a40| 6f 75 20 63 61 6e 20 63 | 72 65 61 74 65 20 61 20 |ou can c|reate a |
|00003a50| 70 72 65 64 69 63 74 69 | 6f 6e 20 6d 6f 64 65 6c |predicti|on model|
|00003a60| 20 28 6c 69 6e 2e 20 73 | 79 73 74 65 6d 20 6f 66 | (lin. s|ystem of|
|00003a70| 20 65 71 75 61 74 69 6f | 6e 73 20 65 2e 67 2e 29 | equatio|ns e.g.)|
|00003a80| 20 63 6f 6e 73 69 64 65 | 72 69 6e 67 20 34 30 20 | conside|ring 40 |
|00003a90| 76 61 72 69 61 62 6c 65 | 73 20 62 75 74 20 6f 6e |variable|s but on|
|00003aa0| 6c 79 20 33 30 20 6f 62 | 73 65 72 76 61 74 69 6f |ly 30 ob|servatio|
|00003ab0| 6e 73 20 66 6f 72 20 65 | 61 63 68 20 76 61 72 69 |ns for e|ach vari|
|00003ac0| 61 62 6c 65 2e 20 59 6f | 75 20 63 61 6e 20 63 6f |able. Yo|u can co|
|00003ad0| 6e 73 69 64 65 72 20 75 | 70 20 74 6f 20 35 30 30 |nsider u|p to 500|
|00003ae0| 20 69 6e 70 75 74 20 76 | 61 72 69 61 62 6c 65 73 | input v|ariables|
|00003af0| 20 28 74 69 6d 65 20 6c | 61 67 67 65 64 20 61 6e | (time l|agged an|
|00003b00| 64 20 75 6e 6c 61 67 67 | 65 64 29 20 69 6e 20 4b |d unlagg|ed) in K|
|00003b10| 6e 6f 77 6c 65 64 67 65 | 4d 69 6e 65 72 20 70 72 |nowledge|Miner pr|
|00003b20| 6f 20 74 6f 20 6d 6f 64 | 65 6c 20 63 6f 6d 70 6c |o to mod|el compl|
|00003b30| 65 78 20 74 69 6d 65 20 | 70 72 6f 63 65 73 73 65 |ex time |processe|
|00003b40| 73 2e 20 41 64 64 69 74 | 69 6f 6e 61 6c 6c 79 2c |s. Addit|ionally,|
|00003b50| 20 4b 6e 6f 77 6c 65 64 | 67 65 4d 69 6e 65 72 20 | Knowled|geMiner |
|00003b60| 70 72 6f 20 68 61 73 20 | 69 6d 70 6c 65 6d 65 6e |pro has |implemen|
|00003b70| 74 65 64 20 74 68 65 20 | 41 6e 61 6c 6f 67 20 43 |ted the |Analog C|
|00003b80| 6f 6d 70 6c 65 78 69 6e | 67 20 6d 65 74 68 6f 64 |omplexin|g method|
|00003b90| 20 61 73 20 61 20 73 65 | 63 6f 6e 64 20 65 78 74 | as a se|cond ext|
|00003ba0| 72 65 6d 65 6c 79 20 70 | 6f 77 65 72 66 75 6c 20 |remely p|owerful |
|00003bb0| 70 72 65 64 69 63 74 69 | 6f 6e 20 74 65 63 68 6e |predicti|on techn|
|00003bc0| 69 71 75 65 20 66 6f 72 | 20 66 75 7a 7a 79 20 70 |ique for| fuzzy p|
|00003bd0| 72 6f 63 65 73 73 65 73 | 20 6c 69 6b 65 20 66 69 |rocesses| like fi|
|00003be0| 6e 61 6e 63 69 61 6c 20 | 6d 61 72 6b 65 74 73 2e |nancial |markets.|
|00003bf0| 20 4b 6e 6f 77 6c 65 64 | 67 65 4d 69 6e 65 72 20 | Knowled|geMiner |
|00003c00| 77 68 65 6e 20 75 73 65 | 64 20 6f 6e 20 66 69 6e |when use|d on fin|
|00003c10| 61 6e 63 69 61 6c 20 6d | 61 72 6b 65 74 73 20 63 |ancial m|arkets c|
|00003c20| 6f 75 6c 64 20 72 65 61 | 6c 6c 79 20 73 74 72 69 |ould rea|lly stri|
|00003c30| 6b 65 20 67 6f 6c 64 21 | 0d 0d 51 3a 20 57 68 65 |ke gold!|..Q: Whe|
|00003c40| 6e 20 79 6f 75 20 73 61 | 79 20 4b 6e 6f 77 6c 65 |n you sa|y Knowle|
|00003c50| 64 67 65 4d 69 6e 65 72 | 20 4c 69 74 65 20 69 73 |dgeMiner| Lite is|
|00003c60| 20 6c 69 6d 69 74 65 64 | 20 74 6f 20 35 30 20 69 | limited| to 50 i|
|00003c70| 6e 70 75 74 73 2c 20 64 | 6f 65 73 20 74 68 61 74 |nputs, d|oes that|
|00003c80| 20 6d 65 61 6e 20 35 30 | 20 72 6f 77 73 20 6f 66 | mean 50| rows of|
|00003c90| 20 64 61 74 61 3f 20 46 | 6f 72 20 65 78 61 6d 70 | data? F|or examp|
|00003ca0| 6c 65 2c 20 61 73 73 75 | 6d 65 20 49 20 77 61 6e |le, assu|me I wan|
|00003cb0| 74 65 64 20 74 6f 20 6d | 6f 64 65 6c 20 74 68 65 |ted to m|odel the|
|00003cc0| 20 72 65 75 6c 74 73 20 | 6f 66 20 35 20 76 61 72 | reults |of 5 var|
|00003cd0| 69 61 62 6c 65 73 20 65 | 78 70 72 65 73 73 65 64 |iables e|xpressed|
|00003ce0| 20 61 73 20 63 6f 6c 75 | 6d 6e 73 2c 20 63 61 6e | as colu|mns, can|
|00003cf0| 20 49 20 68 61 76 65 20 | 35 30 20 72 6f 77 73 20 | I have |50 rows |
|00003d00| 6f 66 20 64 61 74 61 20 | 77 69 74 68 20 35 20 76 |of data |with 5 v|
|00003d10| 61 72 69 61 62 6c 65 73 | 20 69 6e 20 65 61 63 68 |ariables| in each|
|00003d20| 20 72 6f 77 2c 20 6f 72 | 20 61 6d 20 49 20 6c 69 | row, or| am I li|
|00003d30| 6d 69 74 65 64 20 74 6f | 20 31 30 20 72 6f 77 73 |mited to| 10 rows|
|00003d40| 20 6f 66 20 64 61 74 61 | 20 28 31 30 20 72 6f 77 | of data| (10 row|
|00003d50| 73 20 74 69 6d 65 73 20 | 35 20 76 61 72 69 61 62 |s times |5 variab|
|00003d60| 6c 65 73 20 65 71 75 61 | 6c 73 20 35 30 20 76 61 |les equa|ls 50 va|
|00003d70| 72 69 61 62 6c 65 73 20 | 6f 72 20 69 6e 70 75 74 |riables |or input|
|00003d80| 73 29 3f 20 49 20 61 6d | 20 72 65 61 6c 6c 79 20 |s)? I am| really |
|00003d90| 63 6f 6e 66 75 73 65 64 | 20 61 62 6f 75 74 20 74 |confused| about t|
|00003da0| 68 65 20 6c 69 6d 69 74 | 61 74 69 6f 6e 20 6f 66 |he limit|ation of|
|00003db0| 20 35 30 20 69 6e 70 75 | 74 73 2e 0d 41 3a 20 59 | 50 inpu|ts..A: Y|
|00003dc0| 6f 75 20 61 72 65 20 6e | 6f 74 20 6c 69 6d 69 74 |ou are n|ot limit|
|00003dd0| 65 64 20 74 6f 20 35 30 | 20 72 6f 77 73 2e 20 57 |ed to 50| rows. W|
|00003de0| 69 74 68 20 35 30 20 69 | 6e 70 75 74 73 20 61 72 |ith 50 i|nputs ar|
|00003df0| 65 20 6d 65 61 6e 74 20 | 69 6e 70 75 74 20 76 61 |e meant |input va|
|00003e00| 72 69 61 62 6c 65 73 2e | 20 46 6f 72 20 73 74 61 |riables.| For sta|
|00003e10| 74 69 63 20 6d 6f 64 65 | 6c 73 20 77 68 69 63 68 |tic mode|ls which|
|00003e20| 20 63 6f 6e 74 61 69 6e | 20 6e 6f 20 74 69 6d 65 | contain| no time|
|00003e30| 20 6c 61 67 67 65 64 20 | 76 61 72 69 61 62 6c 65 | lagged |variable|
|00003e40| 73 20 69 74 20 69 73 20 | 65 71 75 61 6c 20 74 6f |s it is |equal to|
|00003e50| 20 35 30 20 63 6f 6c 75 | 6d 6e 73 2e 20 46 6f 72 | 50 colu|mns. For|
|00003e60| 20 64 79 6e 61 6d 69 63 | 20 6d 6f 64 65 6c 73 20 | dynamic| models |
|00003e70| 75 73 65 64 20 74 6f 20 | 72 65 66 6c 65 63 74 20 |used to |reflect |
|00003e80| 74 69 6d 65 20 70 72 6f | 63 65 73 73 65 73 20 69 |time pro|cesses i|
|00003e90| 74 20 69 6e 63 6c 75 64 | 65 73 20 74 68 65 20 74 |t includ|es the t|
|00003ea0| 69 6d 65 20 6c 61 67 67 | 65 64 20 76 61 72 69 61 |ime lagg|ed varia|
|00003eb0| 62 6c 65 73 20 61 6c 73 | 6f 2e 20 53 6f 2c 20 74 |bles als|o. So, t|
|00003ec0| 68 65 20 6d 61 78 69 6d | 75 6d 20 64 61 74 61 20 |he maxim|um data |
|00003ed0| 62 61 73 69 73 20 79 6f | 75 20 63 61 6e 20 75 73 |basis yo|u can us|
|00003ee0| 65 20 69 6e 20 74 68 65 | 20 4c 69 74 65 20 76 65 |e in the| Lite ve|
|00003ef0| 72 73 69 6f 6e 20 69 73 | 20 35 30 20 63 6f 6c 75 |rsion is| 50 colu|
|00003f00| 6d 6e 73 20 74 69 6d 65 | 73 20 33 30 30 20 72 6f |mns time|s 300 ro|
|00003f10| 77 73 20 3d 20 31 35 2c | 30 30 30 20 64 61 74 61 |ws = 15,|000 data|
|00003f20| 20 70 6f 69 6e 74 73 2e | 20 54 68 65 20 6c 69 74 | points.| The lit|
|00003f30| 65 20 76 65 72 73 69 6f | 6e 20 69 73 20 61 64 65 |e versio|n is ade|
|00003f40| 71 75 61 74 65 20 66 6f | 72 20 74 68 65 20 65 78 |quate fo|r the ex|
|00003f50| 61 6d 70 6c 65 20 79 6f | 75 20 6d 65 6e 74 69 6f |ample yo|u mentio|
|00003f60| 6e 65 64 2e 20 54 68 65 | 20 6d 61 78 69 6d 75 6d |ned. The| maximum|
|00003f70| 20 64 61 74 61 20 62 61 | 73 69 73 20 66 6f 72 20 | data ba|sis for |
|00003f80| 74 68 65 20 70 72 6f 20 | 76 65 72 73 69 6f 6e 20 |the pro |version |
|00003f90| 69 73 20 35 30 30 20 69 | 6e 70 75 74 20 76 61 72 |is 500 i|nput var|
|00003fa0| 69 61 62 6c 65 73 20 74 | 69 6d 65 73 20 31 30 30 |iables t|imes 100|
|00003fb0| 30 20 72 6f 77 73 2e 20 | 54 68 69 73 20 69 73 20 |0 rows. |This is |
|00003fc0| 63 6f 6d 66 6f 72 74 61 | 62 6c 65 20 66 6f 72 20 |comforta|ble for |
|00003fd0| 61 6c 6d 6f 73 74 20 61 | 6c 6c 20 70 72 6f 62 6c |almost a|ll probl|
|00003fe0| 65 6d 73 20 73 69 6e 63 | 65 20 4b 6e 6f 77 6c 65 |ems sinc|e Knowle|
|00003ff0| 64 67 65 4d 69 6e 65 72 | 20 64 6f 6e 27 74 20 6e |dgeMiner| don't n|
|00004000| 65 65 64 20 6e 65 61 72 | 6c 79 20 61 73 20 6d 61 |eed near|ly as ma|
|00004010| 6e 79 20 63 61 73 65 73 | 20 61 73 20 4e 4e 27 73 |ny cases| as NN's|
|00004020| 20 6f 72 20 73 74 61 74 | 69 73 74 69 63 73 2e 0d | or stat|istics..|
|00004030| 0d 57 65 20 63 61 6e 20 | 63 72 65 61 74 65 20 63 |.We can |create c|
|00004040| 75 73 74 6f 6d 20 76 65 | 72 73 69 6f 6e 73 20 74 |ustom ve|rsions t|
|00004050| 6f 20 6d 65 65 74 20 61 | 6c 6d 6f 73 74 20 61 6e |o meet a|lmost an|
|00004060| 79 20 6d 6f 64 65 6c 69 | 6e 67 20 6e 65 65 64 2e |y modeli|ng need.|
|00004070| 0d 0d 51 3a 20 48 6f 77 | 20 6d 61 6e 79 20 72 6f |..Q: How| many ro|
|00004080| 77 73 20 28 73 75 62 6a | 65 63 74 73 29 20 61 72 |ws (subj|ects) ar|
|00004090| 65 20 6e 65 63 65 73 73 | 61 72 79 20 70 65 72 20 |e necess|ary per |
|000040a0| 63 6f 6c 75 6d 6e 73 20 | 28 76 61 72 69 61 62 6c |columns |(variabl|
|000040b0| 65 73 29 20 74 6f 20 61 | 73 73 75 72 65 20 76 61 |es) to a|ssure va|
|000040c0| 6c 69 64 69 74 79 20 6f | 66 20 74 68 65 20 6d 6f |lidity o|f the mo|
|000040d0| 64 65 6c 2e 20 4f 66 20 | 63 6f 75 72 73 65 20 64 |del. Of |course d|
|000040e0| 75 65 20 74 6f 20 6c 69 | 6d 69 74 61 74 69 6f 6e |ue to li|mitation|
|000040f0| 20 6f 66 20 74 68 65 20 | 64 65 6d 6f 20 76 65 72 | of the |demo ver|
|00004100| 73 69 6f 6e 20 28 6c 69 | 6d 69 74 61 74 69 6f 6e |sion (li|mitation|
|00004110| 20 6f 6e 20 74 68 65 20 | 72 61 77 73 20 74 61 6b | on the |raws tak|
|00004120| 65 6e 20 69 6e 74 6f 20 | 61 63 63 6f 75 6e 74 29 |en into |account)|
|00004130| 2c 20 74 68 65 20 66 69 | 74 74 65 64 20 76 61 6c |, the fi|tted val|
|00004140| 75 65 73 20 77 65 72 65 | 20 70 65 72 66 65 63 74 |ues were| perfect|
|00004150| 73 2c 20 62 75 74 20 70 | 72 65 64 69 63 74 65 64 |s, but p|redicted|
|00004160| 20 76 61 6c 75 65 73 20 | 6f 66 20 74 68 65 20 6d | values |of the m|
|00004170| 6f 64 65 6c 20 77 65 72 | 65 20 77 6f 72 73 65 20 |odel wer|e worse |
|00004180| 6f 6e 20 6d 79 20 70 65 | 72 73 6f 6e 61 6c 20 64 |on my pe|rsonal d|
|00004190| 61 74 61 2e 0d 41 3a 20 | 47 65 6e 65 72 61 6c 6c |ata..A: |Generall|
|000041a0| 79 2c 20 74 68 65 72 65 | 20 69 73 20 6e 6f 20 72 |y, there| is no r|
|000041b0| 75 6c 65 20 6f 6e 20 68 | 6f 77 20 6d 61 6e 79 20 |ule on h|ow many |
|000041c0| 63 61 73 65 73 20 28 72 | 6f 77 73 29 20 61 72 65 |cases (r|ows) are|
|000041d0| 20 6e 65 63 65 73 73 61 | 72 79 20 74 6f 20 6d 6f | necessa|ry to mo|
|000041e0| 64 65 6c 20 61 20 73 70 | 65 63 69 66 69 63 20 6f |del a sp|ecific o|
|000041f0| 62 6a 65 63 74 20 61 70 | 70 72 6f 70 72 69 61 74 |bject ap|propriat|
|00004200| 65 6c 79 2e 20 54 68 65 | 20 72 65 61 73 6f 6e 20 |ely. The| reason |
|00004210| 69 73 20 74 68 61 74 20 | 77 65 20 63 61 6e 20 6f |is that |we can o|
|00004220| 6e 6c 79 20 64 65 61 6c | 20 77 69 74 68 20 61 20 |nly deal| with a |
|00004230| 66 69 6e 69 74 65 20 6e | 75 6d 62 65 72 20 6f 66 |finite n|umber of|
|00004240| 20 64 61 74 61 20 77 68 | 69 6c 65 20 6f 75 72 20 | data wh|ile our |
|00004250| 72 65 61 6c 20 77 6f 72 | 6c 64 20 69 73 20 69 6e |real wor|ld is in|
|00004260| 66 69 6e 69 74 65 2e 0d | 0d 54 68 65 20 6d 6f 64 |finite..|.The mod|
|00004270| 65 6c 69 6e 67 20 6d 65 | 74 68 6f 64 20 69 6d 70 |eling me|thod imp|
|00004280| 6c 65 6d 65 6e 74 65 64 | 20 69 6e 20 4b 6e 6f 77 |lemented| in Know|
|00004290| 6c 65 64 67 65 4d 69 6e | 65 72 20 77 61 73 20 64 |ledgeMin|er was d|
|000042a0| 65 73 69 67 6e 65 64 20 | 74 6f 20 77 6f 72 6b 20 |esigned |to work |
|000042b0| 6f 6e 20 73 6d 61 6c 6c | 20 61 6e 64 20 6e 6f 69 |on small| and noi|
|000042c0| 73 79 20 64 61 74 61 20 | 73 65 74 73 20 61 6e 64 |sy data |sets and|
|000042d0| 20 74 6f 20 66 69 6e 64 | 20 61 20 6d 6f 64 65 6c | to find| a model|
|000042e0| 20 74 68 61 74 20 72 65 | 66 6c 65 63 74 73 20 73 | that re|flects s|
|000042f0| 69 67 6e 69 66 69 63 61 | 6e 74 20 72 65 6c 61 74 |ignifica|nt relat|
|00004300| 69 6f 6e 73 68 69 70 73 | 20 62 65 74 77 65 65 6e |ionships| between|
|00004310| 20 76 61 72 69 61 62 6c | 65 73 2e 20 42 75 74 20 | variabl|es. But |
|00004320| 74 68 65 20 6d 6f 72 65 | 20 61 6e 20 6f 62 6a 65 |the more| an obje|
|00004330| 63 74 20 63 61 6e 20 76 | 61 72 79 20 28 74 68 65 |ct can v|ary (the|
|00004340| 20 6d 6f 72 65 20 73 69 | 67 6e 69 66 69 63 61 6e | more si|gnifican|
|00004350| 74 6c 79 20 64 69 66 66 | 65 72 65 6e 74 20 63 61 |tly diff|erent ca|
|00004360| 73 65 73 20 6f 72 20 73 | 75 62 6a 65 63 74 73 20 |ses or s|ubjects |
|00004370| 69 74 20 63 61 6e 20 68 | 61 76 65 29 20 74 68 65 |it can h|ave) the|
|00004380| 20 6d 6f 72 65 20 64 61 | 74 61 20 28 72 6f 77 73 | more da|ta (rows|
|00004390| 29 20 6f 6e 65 20 73 68 | 6f 75 6c 64 20 75 73 65 |) one sh|ould use|
|000043a0| 20 66 6f 72 20 74 72 61 | 69 6e 69 6e 67 20 61 20 | for tra|ining a |
|000043b0| 6d 6f 64 65 6c 2e 20 41 | 64 64 69 74 69 6f 6e 61 |model. A|dditiona|
|000043c0| 6c 6c 79 2c 20 6e 6f 6e | 6c 69 6e 65 61 72 20 6d |lly, non|linear m|
|000043d0| 6f 64 65 6c 73 20 73 68 | 6f 75 6c 64 20 62 65 20 |odels sh|ould be |
|000043e0| 63 72 65 61 74 65 64 20 | 6f 6e 20 20 61 20 6c 61 |created |on a la|
|000043f0| 72 67 65 72 20 73 65 74 | 20 74 68 61 6e 20 6c 69 |rger set| than li|
|00004400| 6e 65 61 72 20 6d 6f 64 | 65 6c 73 2e 20 45 78 70 |near mod|els. Exp|
|00004410| 65 72 69 65 6e 63 65 20 | 68 61 73 20 73 68 6f 77 |erience |has show|
|00004420| 6e 20 74 68 61 74 20 6c | 69 6e 65 61 72 20 6d 6f |n that l|inear mo|
|00004430| 64 65 6c 73 20 63 72 65 | 61 74 65 64 20 6f 6e 20 |dels cre|ated on |
|00004440| 31 30 20 72 6f 77 73 20 | 6f 66 20 64 61 74 61 20 |10 rows |of data |
|00004450| 63 61 6e 20 70 65 72 66 | 6f 72 6d 20 76 65 72 79 |can perf|orm very|
|00004460| 20 77 65 6c 6c 20 6f 6e | 20 6e 65 77 20 64 61 74 | well on| new dat|
|00004470| 61 20 61 6c 72 65 61 64 | 79 2e 20 49 74 73 20 62 |a alread|y. Its b|
|00004480| 65 74 74 65 72 20 74 6f | 20 75 73 65 20 32 30 20 |etter to| use 20 |
|00004490| 6f 72 20 33 30 20 72 6f | 77 73 20 61 73 20 61 20 |or 30 ro|ws as a |
|000044a0| 6c 6f 77 65 72 20 6c 69 | 6d 69 74 2e 20 46 6f 72 |lower li|mit. For|
|000044b0| 20 6e 6f 6e 6c 69 6e 65 | 61 72 20 6d 6f 64 65 6c | nonline|ar model|
|000044c0| 73 20 61 62 6f 75 74 20 | 35 30 20 72 6f 77 73 20 |s about |50 rows |
|000044d0| 61 72 65 20 61 64 76 69 | 73 61 62 6c 65 20 61 73 |are advi|sable as|
|000044e0| 20 61 20 6c 6f 77 65 72 | 20 6c 69 6d 69 74 20 62 | a lower| limit b|
|000044f0| 75 74 20 69 74 27 73 20 | 6e 6f 74 20 61 20 6d 75 |ut it's |not a mu|
|00004500| 73 74 2e 0d 0d 51 3a 20 | 54 68 65 20 70 6f 73 73 |st...Q: |The poss|
|00004510| 69 62 69 6c 69 74 79 20 | 6f 66 20 74 69 6d 65 20 |ibility |of time |
|00004520| 6c 61 67 20 6d 6f 64 65 | 6c 20 69 73 20 72 65 61 |lag mode|l is rea|
|00004530| 6c 6c 79 20 69 6e 74 65 | 72 65 73 74 69 6e 67 20 |lly inte|resting |
|00004540| 74 6f 6f 2e 20 49 6e 20 | 68 75 6d 61 6e 20 74 72 |too. In |human tr|
|00004550| 61 69 6e 69 6e 67 20 73 | 74 75 64 69 65 73 2c 20 |aining s|tudies, |
|00004560| 74 68 65 20 6e 75 6d 62 | 65 72 20 6f 66 20 6d 65 |the numb|er of me|
|00004570| 61 73 75 72 65 73 20 70 | 65 72 20 79 65 61 72 20 |asures p|er year |
|00004580| 69 73 20 76 65 72 79 20 | 6c 6f 77 20 28 32 2d 36 |is very |low (2-6|
|00004590| 29 2c 20 63 6f 6d 70 61 | 72 65 20 74 6f 20 74 65 |), compa|re to te|
|000045a0| 73 74 69 6e 67 20 76 61 | 72 69 61 62 6c 65 73 20 |sting va|riables |
|000045b0| 28 31 30 2d 32 30 29 2e | 20 48 6f 77 20 6d 61 6e |(10-20).| How man|
|000045c0| 79 20 73 75 62 6a 65 63 | 74 73 20 61 72 65 20 6e |y subjec|ts are n|
|000045d0| 65 63 65 73 73 61 72 79 | 20 74 6f 6f 3f 0d 41 3a |ecessary| too?.A:|
|000045e0| 20 54 68 65 20 73 61 6d | 65 20 69 73 20 74 72 75 | The sam|e is tru|
|000045f0| 65 20 69 66 20 79 6f 75 | 20 77 61 6e 74 20 74 6f |e if you| want to|
|00004600| 20 63 72 65 61 74 65 20 | 61 20 64 79 6e 61 6d 69 | create |a dynami|
|00004610| 63 20 6d 6f 64 65 6c 2e | 20 49 6e 20 63 6f 6e 74 |c model.| In cont|
|00004620| 72 61 73 74 20 74 6f 20 | 73 74 61 74 69 73 74 69 |rast to |statisti|
|00004630| 63 73 20 6f 72 20 4e 65 | 75 72 61 6c 20 4e 65 74 |cs or Ne|ural Net|
|00004640| 77 6f 72 6b 73 20 4b 6e | 6f 77 6c 65 64 67 65 4d |works Kn|owledgeM|
|00004650| 69 6e 65 72 20 63 61 6e | 20 64 65 61 6c 20 77 69 |iner can| deal wi|
|00004660| 74 68 20 61 20 73 6d 61 | 6c 6c 65 72 20 6e 75 6d |th a sma|ller num|
|00004670| 62 65 72 20 6f 66 20 63 | 61 73 65 73 20 74 68 61 |ber of c|ases tha|
|00004680| 6e 20 76 61 72 69 61 62 | 6c 65 73 20 28 73 6f 2d |n variab|les (so-|
|00004690| 63 61 6c 6c 65 64 20 75 | 6e 64 65 72 2d 64 65 74 |called u|nder-det|
|000046a0| 65 72 6d 69 6e 65 64 20 | 74 61 73 6b 73 29 2e 20 |ermined |tasks). |
|000046b0| 53 6f 20 69 74 20 69 73 | 20 72 65 61 6c 6c 79 20 |So it is| really |
|000046c0| 70 6f 73 73 69 62 6c 65 | 20 66 6f 72 20 79 6f 75 |possible| for you|
|000046d0| 20 74 6f 20 75 73 65 20 | 32 30 20 76 61 72 69 61 | to use |20 varia|
|000046e0| 62 6c 65 73 20 61 6e 64 | 20 31 32 20 6f 72 20 31 |bles and| 12 or 1|
|000046f0| 38 20 72 6f 77 73 20 6f | 6e 6c 79 20 66 6f 72 20 |8 rows o|nly for |
|00004700| 63 72 65 61 74 69 6f 6e | 20 6f 66 20 61 20 6c 69 |creation| of a li|
|00004710| 6e 65 61 72 20 73 79 73 | 74 65 6d 20 6f 66 20 65 |near sys|tem of e|
|00004720| 71 75 61 74 69 6f 6e 73 | 2e 0d 0d 51 3a 20 49 20 |quations|...Q: I |
|00004730| 68 61 76 65 20 61 20 6f | 6e 65 2d 74 69 6d 65 20 |have a o|ne-time |
|00004740| 70 72 6f 62 6c 65 6d 20 | 69 6e 76 6f 6c 76 69 6e |problem |involvin|
|00004750| 67 20 74 68 65 20 70 72 | 69 63 65 20 61 6e 61 6c |g the pr|ice anal|
|00004760| 79 73 69 73 20 6f 66 20 | 61 62 6f 75 74 20 32 30 |ysis of |about 20|
|00004770| 30 2c 30 30 30 20 76 65 | 68 69 63 6c 65 73 2c 20 |0,000 ve|hicles, |
|00004780| 61 6e 64 20 74 68 65 20 | 66 61 63 74 6f 72 73 20 |and the |factors |
|00004790| 74 68 61 74 20 69 6e 66 | 6c 75 65 6e 63 65 20 70 |that inf|luence p|
|000047a0| 72 69 63 65 2e 20 49 20 | 68 61 76 65 20 69 64 65 |rice. I |have ide|
|000047b0| 6e 74 69 66 69 65 64 20 | 31 37 20 64 65 73 63 72 |ntified |17 descr|
|000047c0| 69 70 74 6f 72 73 20 28 | 66 61 63 74 6f 72 73 20 |iptors (|factors |
|000047d0| 74 68 61 74 20 6d 61 79 | 20 69 6e 66 6c 75 65 6e |that may| influen|
|000047e0| 63 65 20 70 72 69 63 65 | 29 2e 20 49 20 63 61 6e |ce price|). I can|
|000047f0| 20 70 72 6f 62 61 62 6c | 79 20 65 6c 69 6d 69 6e | probabl|y elimin|
|00004800| 61 74 65 20 34 20 74 6f | 20 36 20 6f 66 20 74 68 |ate 4 to| 6 of th|
|00004810| 65 73 65 20 66 61 63 74 | 6f 72 73 20 69 66 20 6e |ese fact|ors if n|
|00004820| 65 63 65 73 73 61 72 79 | 2e 0d 41 3a 20 31 37 20 |ecessary|..A: 17 |
|00004830| 64 65 73 63 72 69 70 74 | 6f 72 73 20 61 72 65 20 |descript|ors are |
|00004840| 6f 6b 2e 20 54 68 65 79 | 20 61 72 65 20 6e 6f 20 |ok. They| are no |
|00004850| 70 72 6f 62 6c 65 6d 2e | 20 42 75 74 20 69 74 20 |problem.| But it |
|00004860| 6d 75 73 74 20 6e 6f 74 | 20 6f 6e 6c 79 20 62 65 |must not| only be|
|00004870| 20 61 20 6f 6e 65 2d 74 | 69 6d 65 20 70 72 6f 62 | a one-t|ime prob|
|00004880| 6c 65 6d 20 73 69 6e 63 | 65 20 74 68 65 20 70 72 |lem sinc|e the pr|
|00004890| 69 63 69 6e 67 20 6d 6f | 64 65 6c 20 6d 61 79 20 |icing mo|del may |
|000048a0| 63 68 61 6e 67 65 20 6f | 76 65 72 20 74 69 6d 65 |change o|ver time|
|000048b0| 2e 20 55 73 69 6e 67 20 | 4b 6e 6f 77 6c 65 64 67 |. Using |Knowledg|
|000048c0| 65 4d 69 6e 65 72 20 79 | 6f 75 20 63 61 6e 20 61 |eMiner y|ou can a|
|000048d0| 6c 77 61 79 73 20 68 61 | 76 65 20 75 70 2d 74 6f |lways ha|ve up-to|
|000048e0| 2d 64 61 74 65 20 6d 6f | 64 65 6c 73 2e 0d 0d 51 |-date mo|dels...Q|
|000048f0| 3a 20 57 68 61 74 20 6b | 69 6e 64 20 6f 66 20 52 |: What k|ind of R|
|00004900| 41 4d 20 69 73 20 6e 65 | 65 64 65 64 20 74 6f 20 |AM is ne|eded to |
|00004910| 75 73 65 20 4b 6e 6f 77 | 6c 65 64 67 65 4d 69 6e |use Know|ledgeMin|
|00004920| 65 72 3f 20 0d 41 3a 20 | 46 6f 72 20 6d 6f 64 65 |er? .A: |For mode|
|00004930| 6c 69 6e 67 20 6f 66 20 | 6c 61 72 67 65 72 20 70 |ling of |larger p|
|00004940| 72 6f 62 6c 65 6d 73 20 | 61 20 6c 6f 74 20 6f 66 |roblems |a lot of|
|00004950| 20 6d 65 6d 6f 72 79 20 | 73 70 61 63 65 20 69 73 | memory |space is|
|00004960| 20 6e 65 65 64 65 64 2e | 20 54 68 65 72 65 66 6f | needed.| Therefo|
|00004970| 72 65 20 4b 6e 6f 77 6c | 65 64 67 65 4d 69 6e 65 |re Knowl|edgeMine|
|00004980| 72 20 75 73 65 73 20 74 | 65 6d 70 6f 72 61 72 69 |r uses t|emporari|
|00004990| 6c 79 2c 20 66 6f 72 20 | 6d 65 6d 6f 72 79 20 69 |ly, for |memory i|
|000049a0| 6e 74 65 6e 73 69 76 65 | 20 74 61 73 6b 73 2c 20 |ntensive| tasks, |
|000049b0| 74 68 65 20 66 72 65 65 | 20 52 41 4d 20 6f 6e 20 |the free| RAM on |
|000049c0| 79 6f 75 72 20 63 6f 6d | 70 75 74 65 72 2e 20 54 |your com|puter. T|
|000049d0| 68 65 20 6d 6f 64 65 6c | 69 6e 67 20 70 72 6f 63 |he model|ing proc|
|000049e0| 65 73 73 20 69 73 20 6d | 65 6d 6f 72 79 20 69 6e |ess is m|emory in|
|000049f0| 74 65 6e 73 69 76 65 20 | 62 65 63 61 75 73 65 20 |tensive |because |
|00004a00| 69 74 20 77 6f 72 6b 73 | 20 6c 69 6b 65 20 6e 61 |it works| like na|
|00004a10| 74 75 72 61 6c 20 65 76 | 6f 6c 75 74 69 6f 6e 3a |tural ev|olution:|
|00004a20| 20 61 20 67 69 76 65 6e | 20 67 65 6e 65 72 61 74 | a given| generat|
|00004a30| 69 6f 6e 20 6f 66 20 69 | 6e 64 69 76 69 64 75 61 |ion of i|ndividua|
|00004a40| 6c 73 20 28 69 6e 70 75 | 74 20 76 61 72 69 61 62 |ls (inpu|t variab|
|00004a50| 6c 65 73 29 20 63 72 65 | 61 74 65 73 20 61 20 6e |les) cre|ates a n|
|00004a60| 65 77 20 67 65 6e 65 72 | 61 74 69 6f 6e 20 6f 66 |ew gener|ation of|
|00004a70| 20 69 6e 64 69 76 69 64 | 75 61 6c 73 20 62 79 20 | individ|uals by |
|00004a80| 61 6c 6c 20 70 6f 73 73 | 69 62 6c 65 20 70 61 69 |all poss|ible pai|
|00004a90| 72 77 69 73 65 20 63 6f | 6d 62 69 6e 61 74 69 6f |rwise co|mbinatio|
|00004aa0| 6e 73 20 77 68 69 63 68 | 20 69 6e 20 74 75 72 6e |ns which| in turn|
|00004ab0| 20 2d 20 61 66 74 65 72 | 20 73 65 6c 65 63 74 69 | - after| selecti|
|00004ac0| 6f 6e 20 6f 66 20 74 68 | 65 20 62 65 73 74 20 66 |on of th|e best f|
|00004ad0| 69 74 74 65 73 74 20 69 | 6e 64 69 76 69 64 75 61 |ittest i|ndividua|
|00004ae0| 6c 73 20 2d 20 63 72 65 | 61 74 65 20 61 67 61 69 |ls - cre|ate agai|
|00004af0| 6e 20 61 20 6e 65 77 20 | 67 65 6e 65 72 61 74 69 |n a new |generati|
|00004b00| 6f 6e 20 74 68 61 74 20 | 62 65 74 74 65 72 20 66 |on that |better f|
|00004b10| 69 74 73 20 74 68 65 20 | 64 65 73 69 72 65 64 20 |its the |desired |
|00004b20| 62 65 68 61 76 69 6f 72 | 20 74 68 61 6e 20 74 68 |behavior| than th|
|00004b30| 65 20 70 72 65 76 69 6f | 75 73 20 61 6e 64 20 73 |e previo|us and s|
|00004b40| 6f 20 6f 6e 2e 20 54 68 | 69 73 20 70 72 6f 63 65 |o on. Th|is proce|
|00004b50| 73 73 69 6e 67 20 72 65 | 71 75 69 72 65 73 20 74 |ssing re|quires t|
|00004b60| 68 61 74 20 61 6c 6c 20 | 69 6e 64 69 76 69 64 75 |hat all |individu|
|00004b70| 61 6c 73 20 6f 66 20 61 | 20 67 65 6e 65 72 61 74 |als of a| generat|
|00004b80| 69 6f 6e 20 6d 75 73 74 | 20 6c 69 76 65 20 61 74 |ion must| live at|
|00004b90| 20 61 20 74 69 6d 65 20 | 69 6e 20 74 68 65 20 63 | a time |in the c|
|00004ba0| 6f 6d 70 75 74 65 72 20 | 73 6f 20 74 68 61 74 20 |omputer |so that |
|00004bb0| 74 68 65 79 20 63 61 6e | 20 62 65 20 63 6f 6d 70 |they can| be comp|
|00004bc0| 61 72 65 64 20 61 6e 64 | 20 73 65 6c 65 63 74 65 |ared and| selecte|
|00004bd0| 64 20 6f 75 74 2e 20 54 | 68 65 20 62 61 73 69 63 |d out. T|he basic|
|00004be0| 20 72 75 6c 65 20 66 6f | 72 20 74 68 69 73 20 72 | rule fo|r this r|
|00004bf0| 65 6c 61 74 69 6f 6e 20 | 69 73 3a 20 74 68 65 20 |elation |is: the |
|00004c00| 6d 6f 72 65 20 69 6e 70 | 75 74 73 20 75 73 65 64 |more inp|uts used|
|00004c10| 20 74 68 65 20 6d 6f 72 | 65 20 69 6e 64 69 76 69 | the mor|e indivi|
|00004c20| 64 75 61 6c 73 20 6d 75 | 73 74 20 62 65 20 63 72 |duals mu|st be cr|
|00004c30| 65 61 74 65 64 20 61 6e | 64 20 74 68 65 20 6d 6f |eated an|d the mo|
|00004c40| 72 65 20 6d 65 6d 6f 72 | 79 20 69 73 20 6e 65 65 |re memor|y is nee|
|00004c50| 64 65 64 20 28 6e 6f 6e | 6c 69 6e 65 61 72 20 72 |ded (non|linear r|
|00004c60| 65 6c 61 74 69 6f 6e 73 | 68 69 70 29 2e 0d 0d 51 |elations|hip)...Q|
|00004c70| 3a 20 57 68 61 74 20 77 | 6f 75 6c 64 20 62 65 20 |: What w|ould be |
|00004c80| 74 68 65 20 6c 61 72 67 | 65 73 74 20 74 61 62 6c |the larg|est tabl|
|00004c90| 65 20 28 63 6f 6c 75 6d | 6e 73 20 61 6e 64 20 72 |e (colum|ns and r|
|00004ca0| 6f 77 73 29 0d 4b 6e 6f | 77 6c 65 64 67 65 4d 69 |ows).Kno|wledgeMi|
|00004cb0| 6e 65 72 20 63 6f 75 6c | 64 20 61 63 63 6f 6d 6d |ner coul|d accomm|
|00004cc0| 6f 64 61 74 65 20 69 66 | 20 61 6c 6c 6f 63 61 74 |odate if| allocat|
|00004cd0| 65 64 20 31 30 30 20 4d | 42 20 6f 66 20 52 41 4d |ed 100 M|B of RAM|
|00004ce0| 3f 0d 0d 41 3a 20 54 68 | 65 20 74 61 62 6c 65 20 |?..A: Th|e table |
|00004cf0| 63 6f 6e 74 61 69 6e 73 | 20 61 70 70 72 6f 78 69 |contains| approxi|
|00004d00| 6d 61 74 65 64 20 76 61 | 6c 75 65 73 20 61 73 20 |mated va|lues as |
|00004d10| 61 6e 20 6f 72 69 65 6e | 74 61 74 69 6f 6e 3a 0d |an orien|tation:.|
|00004d20| 20 20 20 20 20 20 20 20 | 20 31 30 30 20 72 6f 77 | | 100 row|
|00004d30| 73 20 20 20 20 20 20 20 | 2d 20 20 20 35 30 30 20 |s |- 500 |
|00004d40| 69 6e 70 75 74 73 0d 20 | 20 20 20 20 20 20 20 20 |inputs. | |
|00004d50| 32 30 30 20 20 20 20 20 | 20 20 20 20 20 20 20 20 |200 | |
|00004d60| 20 20 2d 20 20 20 33 35 | 30 0d 20 20 20 20 20 20 | - 35|0. |
|00004d70| 20 20 20 33 30 30 20 20 | 20 20 20 20 20 20 20 20 | 300 | |
|00004d80| 20 20 20 20 20 2d 20 20 | 20 32 38 30 0d 20 20 20 | - | 280. |
|00004d90| 20 20 20 20 20 20 34 30 | 30 20 20 20 20 20 20 20 | 40|0 |
|00004da0| 20 20 20 20 20 20 20 20 | 2d 20 20 20 32 34 30 0d | |- 240.|
|00004db0| 20 20 20 20 20 20 20 20 | 20 35 30 30 20 20 20 20 | | 500 |
|00004dc0| 20 20 20 20 20 20 20 20 | 20 20 20 2d 20 20 20 32 | | - 2|
|00004dd0| 31 30 20 20 20 20 20 20 | 20 20 20 20 20 20 0d 20 |10 | . |
|00004de0| 20 20 20 20 20 20 20 20 | 31 30 30 30 20 20 20 20 | |1000 |
|00004df0| 20 20 20 20 20 20 20 20 | 20 2d 20 20 20 31 35 30 | | - 150|
|00004e00| 0d 20 20 20 20 20 20 20 | 20 20 32 30 30 30 20 20 |. | 2000 |
|00004e10| 20 20 20 20 20 20 20 20 | 20 20 20 2d 20 20 20 31 | | - 1|
|00004e20| 31 30 0d 20 20 20 20 20 | 20 20 20 20 35 30 30 30 |10. | 5000|
|00004e30| 20 20 20 20 20 20 20 20 | 20 20 20 20 20 2d 20 20 | | - |
|00004e40| 20 20 20 37 30 0d 0d 51 | 3a 20 49 73 20 74 68 65 | 70..Q|: Is the|
|00004e50| 72 65 20 61 20 6c 69 6d | 69 74 20 6f 6e 20 74 68 |re a lim|it on th|
|00004e60| 65 20 6e 75 6d 62 65 72 | 20 6f 66 20 6c 61 79 65 |e number| of laye|
|00004e70| 72 73 20 73 75 62 6a 65 | 63 74 65 64 20 74 6f 20 |rs subje|cted to |
|00004e80| 61 6e 61 6c 79 73 69 73 | 3f 0d 41 3a 20 59 65 73 |analysis|?.A: Yes|
|00004e90| 2c 20 74 68 65 72 65 20 | 69 73 20 61 20 6c 69 6d |, there |is a lim|
|00004ea0| 69 74 2e 20 55 6e 74 69 | 6c 20 6e 6f 77 20 69 74 |it. Unti|l now it|
|00004eb0| 20 77 61 73 20 6e 6f 74 | 20 72 65 6c 65 76 61 6e | was not| relevan|
|00004ec0| 74 20 70 72 61 63 74 69 | 63 61 6c 6c 79 2c 20 65 |t practi|cally, e|
|00004ed0| 76 65 6e 20 66 6f 72 20 | 6c 61 72 67 65 20 73 79 |ven for |large sy|
|00004ee0| 73 74 65 6d 73 20 28 32 | 30 30 20 61 6e 64 20 6d |stems (2|00 and m|
|00004ef0| 6f 72 65 20 69 6e 70 75 | 74 73 29 2e 20 54 68 65 |ore inpu|ts). The|
|00004f00| 20 70 72 6f 63 65 73 73 | 20 61 6c 77 61 79 73 20 | process| always |
|00004f10| 73 74 6f 70 73 20 61 75 | 74 6f 6d 61 74 69 63 61 |stops au|tomatica|
|00004f20| 6c 6c 79 20 28 61 6e 20 | 6f 70 74 69 6d 61 6c 20 |lly (an |optimal |
|00004f30| 63 6f 6d 70 6c 65 78 20 | 6d 6f 64 65 6c 20 77 61 |complex |model wa|
|00004f40| 73 20 66 6f 75 6e 64 29 | 20 62 65 66 6f 72 65 20 |s found)| before |
|00004f50| 74 68 61 74 20 70 68 79 | 73 69 63 61 6c 20 6c 69 |that phy|sical li|
|00004f60| 6d 69 74 2e 0d 0d 51 3a | 20 43 61 6e 20 4b 6e 6f |mit...Q:| Can Kno|
|00004f70| 77 6c 65 64 67 65 4d 69 | 6e 65 72 20 72 65 61 64 |wledgeMi|ner read|
|00004f80| 20 53 41 53 20 66 69 6c | 65 73 2e 0d 41 3a 20 4b | SAS fil|es..A: K|
|00004f90| 6e 6f 77 6c 65 64 67 65 | 4d 69 6e 65 72 20 63 61 |nowledge|Miner ca|
|00004fa0| 6e 6e 6f 74 20 72 65 61 | 64 20 53 41 53 20 66 69 |nnot rea|d SAS fi|
|00004fb0| 6c 65 73 2e 20 42 75 74 | 20 53 41 53 20 63 61 6e |les. But| SAS can|
|00004fc0| 20 73 61 76 65 20 64 61 | 74 61 20 61 73 20 41 53 | save da|ta as AS|
|00004fd0| 43 49 49 20 20 66 69 6c | 65 73 20 61 6e 64 20 74 |CII fil|es and t|
|00004fe0| 68 65 79 20 63 61 6e 20 | 62 65 20 72 65 61 64 20 |hey can |be read |
|00004ff0| 62 79 20 4b 6e 6f 77 6c | 65 64 67 65 4d 69 6e 65 |by Knowl|edgeMine|
|00005000| 72 2e 20 4f 72 20 69 6d | 70 6f 72 74 20 76 69 61 |r. Or im|port via|
|00005010| 20 63 6f 70 79 20 61 6e | 64 20 70 61 73 74 65 2e | copy an|d paste.|
|00005020| 20 57 65 20 72 65 63 6f | 6d 6d 65 6e 64 20 73 70 | We reco|mmend sp|
|00005030| 6c 69 74 74 69 6e 67 20 | 76 65 72 79 20 6c 61 72 |litting |very lar|
|00005040| 67 65 20 44 61 74 61 20 | 62 61 73 65 73 20 6e 65 |ge Data |bases ne|
|00005050| 65 64 65 64 20 74 6f 20 | 64 6f 20 73 74 61 74 69 |eded to |do stati|
|00005060| 73 74 69 63 73 20 28 31 | 30 20 74 68 6f 75 73 61 |stics (1|0 thousa|
|00005070| 6e 64 73 20 6f 66 20 72 | 6f 77 73 20 6f 72 20 73 |nds of r|ows or s|
|00005080| 6f 29 20 69 6e 74 6f 20 | 73 6d 61 6c 6c 65 72 20 |o) into |smaller |
|00005090| 70 72 6f 62 6c 65 6d 73 | 20 77 68 69 63 68 20 69 |problems| which i|
|000050a0| 73 20 61 6c 6d 6f 73 74 | 20 61 6c 77 61 79 73 20 |s almost| always |
|000050b0| 70 6f 73 73 69 62 6c 65 | 2e 20 4c 61 72 67 65 20 |possible|. Large |
|000050c0| 64 61 74 61 20 73 65 74 | 73 20 61 72 65 20 6f 66 |data set|s are of|
|000050d0| 74 65 6e 20 68 69 67 68 | 6c 79 20 72 65 64 75 6e |ten high|ly redun|
|000050e0| 64 61 6e 74 2e 20 41 20 | 6d 75 63 68 20 73 6d 61 |dant. A |much sma|
|000050f0| 6c 6c 65 72 20 64 61 74 | 61 20 73 65 74 20 63 61 |ller dat|a set ca|
|00005100| 6e 20 63 6f 6e 74 61 69 | 6e 20 74 68 65 20 73 61 |n contai|n the sa|
|00005110| 6d 65 20 69 6e 66 6f 72 | 6d 61 74 69 6f 6e 20 61 |me infor|mation a|
|00005120| 73 20 61 20 6d 75 63 68 | 20 6c 61 72 67 65 72 20 |s a much| larger |
|00005130| 6f 6e 65 2e 20 42 79 20 | 74 68 65 20 77 61 79 2c |one. By |the way,|
|00005140| 20 64 6f 77 6e 73 69 7a | 69 6e 67 20 6c 61 72 67 | downsiz|ing larg|
|00005150| 65 20 70 72 6f 62 6c 65 | 6d 73 20 69 73 20 61 20 |e proble|ms is a |
|00005160| 70 72 69 6e 63 69 70 6c | 65 20 4b 6e 6f 77 6c 65 |principl|e Knowle|
|00005170| 64 67 65 4d 69 6e 65 72 | 20 69 73 20 62 75 69 6c |dgeMiner| is buil|
|00005180| 74 20 6f 6e 20 61 6c 73 | 6f 2e 0d 0d 51 3a 20 4f |t on als|o...Q: O|
|00005190| 6e 65 20 6f 66 20 74 68 | 65 20 64 69 66 66 69 63 |ne of th|e diffic|
|000051a0| 75 6c 74 69 65 73 20 77 | 69 74 68 20 53 74 61 74 |ulties w|ith Stat|
|000051b0| 69 73 74 69 63 61 6c 20 | 50 61 74 74 65 72 6e 20 |istical |Pattern |
|000051c0| 52 65 63 6f 67 6e 69 74 | 69 6f 6e 20 69 6e 20 6d |Recognit|ion in m|
|000051d0| 79 20 61 70 70 6c 69 63 | 61 74 69 6f 6e 20 69 73 |y applic|ation is|
|000051e0| 20 74 68 61 74 20 6f 6e | 65 20 6d 69 67 68 74 20 | that on|e might |
|000051f0| 6e 6f 74 20 67 65 74 20 | 61 20 73 75 66 66 69 63 |not get |a suffic|
|00005200| 69 65 6e 74 6c 79 20 73 | 6f 70 68 69 73 74 69 63 |iently s|ophistic|
|00005210| 61 74 65 64 20 63 6c 61 | 73 73 69 66 69 65 72 20 |ated cla|ssifier |
|00005220| 74 6f 20 67 69 76 65 20 | 74 68 65 20 62 65 73 74 |to give |the best|
|00005230| 20 70 6f 73 73 69 62 6c | 65 20 72 65 73 75 6c 74 | possibl|e result|
|00005240| 73 20 28 66 6f 72 20 65 | 78 61 6d 70 6c 65 20 75 |s (for e|xample u|
|00005250| 73 69 6e 67 20 61 20 6c | 69 6e 65 61 72 20 63 6c |sing a l|inear cl|
|00005260| 61 73 73 69 66 69 65 72 | 20 69 6e 73 74 65 61 64 |assifier| instead|
|00005270| 20 6f 66 20 61 20 6d 6f | 72 65 20 63 6f 6d 70 6c | of a mo|re compl|
|00005280| 65 78 20 71 75 61 64 72 | 61 74 69 63 20 63 6c 61 |ex quadr|atic cla|
|00005290| 73 73 69 66 69 65 72 29 | 2e 20 20 49 74 20 61 70 |ssifier)|. It ap|
|000052a0| 70 65 61 72 73 20 74 68 | 61 74 20 4b 4d 20 64 6f |pears th|at KM do|
|000052b0| 65 73 20 6e 6f 74 20 73 | 75 66 66 65 72 20 66 72 |es not s|uffer fr|
|000052c0| 6f 6d 20 74 68 69 73 20 | 70 72 6f 62 6c 65 6d 20 |om this |problem |
|000052d0| 62 65 63 61 75 73 65 20 | 69 73 20 61 70 70 65 61 |because |is appea|
|000052e0| 72 73 20 74 6f 20 70 72 | 6f 64 75 63 65 20 61 20 |rs to pr|oduce a |
|000052f0| 62 6f 6e 61 66 69 64 65 | 20 6e 6f 6e 6c 69 6e 65 |bonafide| nonline|
|00005300| 61 72 20 65 71 75 61 74 | 69 6f 6e 20 77 68 69 63 |ar equat|ion whic|
|00005310| 68 20 73 68 6f 75 6c 64 | 20 6f 70 74 69 6d 61 6c |h should| optimal|
|00005320| 6c 79 20 61 63 63 6f 6d | 6d 6f 64 61 74 65 20 61 |ly accom|modate a|
|00005330| 6e 79 20 69 72 72 65 67 | 75 6c 61 72 20 73 68 61 |ny irreg|ular sha|
|00005340| 70 69 6e 67 20 6f 66 20 | 74 68 65 20 63 6c 61 73 |ping of |the clas|
|00005350| 73 20 70 6f 70 75 6c 61 | 74 69 6f 6e 73 20 69 6e |s popula|tions in|
|00005360| 20 66 65 61 74 75 72 65 | 20 73 70 61 63 65 2e 20 | feature| space. |
|00005370| 20 49 73 20 74 68 69 73 | 20 74 72 75 65 3f 0d 41 | Is this| true?.A|
|00005380| 3a 20 59 65 73 2c 20 79 | 6f 75 20 61 72 65 20 63 |: Yes, y|ou are c|
|00005390| 6f 72 72 65 63 74 2e 20 | 4f 6e 65 20 69 6d 70 6f |orrect. |One impo|
|000053a0| 72 74 61 6e 74 20 66 65 | 61 74 75 72 65 20 6f 66 |rtant fe|ature of|
|000053b0| 20 4b 6e 6f 77 6c 65 64 | 67 65 4d 69 6e 65 72 20 | Knowled|geMiner |
|000053c0| 69 73 20 74 68 61 74 20 | 69 74 20 63 72 65 61 74 |is that |it creat|
|000053d0| 65 73 20 6d 6f 64 65 6c | 73 20 69 6e 20 61 6e 20 |es model|s in an |
|000053e0| 65 76 6f 6c 75 74 69 6f | 6e 61 72 79 20 77 61 79 |evolutio|nary way|
|000053f0| 3a 20 66 72 6f 6d 20 76 | 65 72 79 20 73 69 6d 70 |: from v|ery simp|
|00005400| 6c 65 20 6d 6f 64 65 6c | 73 20 74 6f 20 69 6e 63 |le model|s to inc|
|00005410| 72 65 61 73 69 6e 67 6c | 79 20 6d 6f 72 65 20 63 |reasingl|y more c|
|00005420| 6f 6d 70 6c 65 78 20 6f | 6e 65 73 2e 20 49 74 20 |omplex o|nes. It |
|00005430| 73 74 6f 70 73 20 61 75 | 74 6f 6d 61 74 69 63 61 |stops au|tomatica|
|00005440| 6c 6c 79 20 77 68 65 6e | 20 61 6e 20 6f 70 74 69 |lly when| an opti|
|00005450| 6d 61 6c 20 63 6f 6d 70 | 6c 65 78 20 6d 6f 64 65 |mal comp|lex mode|
|00005460| 6c 20 77 61 73 20 66 6f | 75 6e 64 2e 20 54 68 61 |l was fo|und. Tha|
|00005470| 74 20 69 73 2c 20 77 68 | 65 6e 20 69 74 20 62 65 |t is, wh|en it be|
|00005480| 67 69 6e 73 20 74 6f 20 | 6f 76 65 72 66 69 74 20 |gins to |overfit |
|00005490| 74 68 65 20 64 65 73 69 | 67 6e 20 64 61 74 61 20 |the desi|gn data |
|000054a0| 28 74 68 65 20 64 61 74 | 61 20 75 73 65 64 20 74 |(the dat|a used t|
|000054b0| 6f 20 63 72 65 61 74 65 | 20 72 65 6c 61 74 69 6f |o create| relatio|
|000054c0| 6e 73 68 69 70 73 20 62 | 65 74 77 65 65 6e 20 76 |nships b|etween v|
|000054d0| 61 72 69 61 62 6c 65 73 | 29 2e 0d 0d 51 3a 20 49 |ariables|)...Q: I|
|000054e0| 74 20 22 66 65 65 6c 73 | 22 20 6c 69 6b 65 20 4b |t "feels|" like K|
|000054f0| 6e 6f 77 6c 65 64 67 65 | 4d 69 6e 65 72 20 6d 69 |nowledge|Miner mi|
|00005500| 67 68 74 20 61 73 73 69 | 73 74 20 69 6e 20 64 65 |ght assi|st in de|
|00005510| 74 65 63 74 69 6e 67 20 | 72 65 6c 61 74 69 6f 6e |tecting |relation|
|00005520| 73 68 69 70 73 20 61 6d | 6f 6e 67 20 63 65 72 74 |ships am|ong cert|
|00005530| 61 69 6e 20 70 61 74 69 | 65 6e 74 20 67 72 6f 75 |ain pati|ent grou|
|00005540| 70 73 20 62 79 20 63 6c | 69 6e 69 63 61 6c 20 63 |ps by cl|inical c|
|00005550| 72 69 74 65 72 69 61 20 | 76 73 2e 20 66 6c 75 69 |riteria |vs. flui|
|00005560| 64 20 6d 65 61 73 75 72 | 65 6d 65 6e 74 73 20 74 |d measur|ements t|
|00005570| 68 61 74 20 6d 61 79 20 | 62 65 20 6d 69 73 73 65 |hat may |be misse|
|00005580| 64 20 62 79 20 61 6e 20 | 69 6e 64 69 76 69 64 75 |d by an |individu|
|00005590| 61 6c 2e 20 49 66 20 49 | 20 75 6e 64 65 72 73 74 |al. If I| underst|
|000055a0| 61 6e 64 20 74 68 65 20 | 61 70 70 6c 69 63 61 74 |and the |applicat|
|000055b0| 69 6f 6e 20 6f 66 20 4b | 6e 6f 77 6c 65 64 67 65 |ion of K|nowledge|
|000055c0| 4d 69 6e 65 72 2c 20 49 | 20 62 65 6c 69 65 76 65 |Miner, I| believe|
|000055d0| 20 49 20 73 68 6f 75 6c | 64 20 62 65 20 61 62 6c | I shoul|d be abl|
|000055e0| 65 20 74 6f 20 74 61 6b | 65 20 6f 75 72 20 64 61 |e to tak|e our da|
|000055f0| 74 61 62 61 73 65 20 6f | 66 20 65 79 65 20 66 65 |tabase o|f eye fe|
|00005600| 61 74 75 72 65 73 20 61 | 6c 6f 6e 67 20 77 69 74 |atures a|long wit|
|00005610| 68 20 74 68 65 20 64 69 | 6f 70 74 65 72 20 6d 65 |h the di|opter me|
|00005620| 61 73 75 72 65 6d 65 6e | 74 73 20 66 6f 72 20 65 |asuremen|ts for e|
|00005630| 61 63 68 20 70 61 74 69 | 65 6e 74 20 69 6e 20 74 |ach pati|ent in t|
|00005640| 68 65 20 64 61 74 61 62 | 61 73 65 2c 20 70 6c 75 |he datab|ase, plu|
|00005650| 67 20 74 68 6f 73 65 20 | 69 6e 74 6f 20 4b 6e 6f |g those |into Kno|
|00005660| 77 6c 65 64 67 65 4d 69 | 6e 65 72 20 61 6e 64 20 |wledgeMi|ner and |
|00005670| 74 68 65 6e 20 4b 6e 6f | 77 6c 65 64 67 65 4d 69 |then Kno|wledgeMi|
|00005680| 6e 65 72 20 77 69 6c 6c | 20 64 65 72 69 76 65 20 |ner will| derive |
|00005690| 61 6e 20 65 71 75 61 74 | 69 6f 6e 20 66 6f 72 20 |an equat|ion for |
|000056a0| 63 61 6c 63 75 6c 61 74 | 69 6e 67 20 74 68 65 20 |calculat|ing the |
|000056b0| 64 69 6f 70 74 65 72 20 | 6d 65 61 73 75 72 65 6d |diopter |measurem|
|000056c0| 65 6e 74 20 6f 66 20 61 | 20 70 61 74 69 65 6e 74 |ent of a| patient|
|000056d0| 20 61 73 20 61 20 66 75 | 6e 63 74 69 6f 6e 20 6f | as a fu|nction o|
|000056e0| 66 20 74 68 65 20 70 61 | 74 69 65 6e 74 27 73 20 |f the pa|tient's |
|000056f0| 69 6d 61 67 65 20 66 65 | 61 74 75 72 65 73 2e 20 |image fe|atures. |
|00005700| 20 49 73 20 74 68 69 73 | 20 74 72 75 65 3f 0d 41 | Is this| true?.A|
|00005710| 3a 20 59 65 73 2c 20 65 | 78 61 63 74 6c 79 2e 20 |: Yes, e|xactly. |
|00005720| 54 68 69 73 20 69 73 20 | 73 6f 6d 65 74 68 69 6e |This is |somethin|
|00005730| 67 20 4b 6e 6f 77 6c 65 | 64 67 65 4d 69 6e 65 72 |g Knowle|dgeMiner|
|00005740| 20 63 61 6e 20 64 6f 2e | 0d 0d 51 3a 20 41 72 65 | can do.|..Q: Are|
|00005750| 20 74 68 65 72 65 20 73 | 74 61 74 69 73 74 69 63 | there s|tatistic|
|00005760| 61 6c 20 74 65 73 74 73 | 20 6f 66 20 73 69 67 6e |al tests| of sign|
|00005770| 69 66 69 63 61 6e 63 65 | 20 74 68 61 74 20 63 61 |ificance| that ca|
|00005780| 6e 20 62 65 20 61 70 70 | 6c 69 65 64 20 74 6f 20 |n be app|lied to |
|00005790| 64 65 72 69 76 65 64 20 | 6d 6f 64 65 6c 20 63 6f |derived |model co|
|000057a0| 65 66 66 69 63 69 65 6e | 74 73 2c 20 69 2e 65 2e |efficien|ts, i.e.|
|000057b0| 2c 20 61 20 74 2d 74 65 | 73 74 20 6f 72 20 61 6e |, a t-te|st or an|
|000057c0| 20 46 2d 74 65 73 74 3f | 20 49 20 61 73 6b 20 61 | F-test?| I ask a|
|000057d0| 62 6f 75 74 20 74 68 65 | 20 6c 61 74 74 65 72 20 |bout the| latter |
|000057e0| 62 65 63 61 75 73 65 20 | 6f 6e 65 20 6f 66 20 74 |because |one of t|
|000057f0| 68 65 20 6d 61 6a 6f 72 | 20 63 72 69 74 69 63 69 |he major| critici|
|00005800| 73 6d 73 20 49 27 76 65 | 20 68 65 61 72 64 20 6c |sms I've| heard l|
|00005810| 65 76 69 65 64 20 61 67 | 61 69 6e 73 74 20 6e 65 |evied ag|ainst ne|
|00005820| 75 72 61 6c 20 6e 65 74 | 77 6f 72 6b 20 6d 6f 64 |ural net|work mod|
|00005830| 65 6c 73 20 69 73 20 74 | 68 61 74 20 74 68 65 72 |els is t|hat ther|
|00005840| 65 20 61 72 65 20 6e 6f | 74 20 73 74 61 74 69 73 |e are no|t statis|
|00005850| 74 69 63 61 6c 20 74 65 | 73 74 73 20 74 68 61 74 |tical te|sts that|
|00005860| 20 63 61 6e 20 62 65 20 | 61 70 70 6c 69 65 64 20 | can be |applied |
|00005870| 61 67 61 69 6e 73 74 20 | 74 68 65 20 64 65 72 69 |against |the deri|
|00005880| 76 65 64 20 6d 6f 64 65 | 6c 2e 20 20 48 65 6e 63 |ved mode|l. Henc|
|00005890| 65 2c 20 69 74 20 69 73 | 20 73 6f 72 74 20 6f 66 |e, it is| sort of|
|000058a0| 20 6c 69 6b 65 20 22 65 | 79 65 62 61 6c 6c 69 6e | like "e|yeballin|
|000058b0| 67 22 20 61 20 72 65 67 | 72 65 73 73 69 6f 6e 20 |g" a reg|ression |
|000058c0| 6c 69 6e 65 20 61 6e 64 | 20 68 61 76 69 6e 67 20 |line and| having |
|000058d0| 6e 6f 20 77 61 79 20 74 | 6f 20 64 65 74 65 72 6d |no way t|o determ|
|000058e0| 69 6e 65 20 69 66 20 6d | 79 20 22 65 79 65 62 61 |ine if m|y "eyeba|
|000058f0| 6c 6c 65 64 22 20 6d 6f | 64 65 6c 20 69 73 20 61 |lled" mo|del is a|
|00005900| 6e 79 20 62 65 74 74 65 | 72 20 74 68 61 6e 20 79 |ny bette|r than y|
|00005910| 6f 75 72 20 22 65 79 65 | 62 61 6c 6c 65 64 22 20 |our "eye|balled" |
|00005920| 6d 6f 64 65 6c 2e 0d 41 | 3a 20 20 59 6f 75 72 20 |model..A|: Your |
|00005930| 71 75 65 73 74 69 6f 6e | 20 66 6f 63 75 73 65 73 |question| focuses|
|00005940| 20 74 68 65 20 70 72 6f | 62 6c 65 6d 20 6f 6e 20 | the pro|blem on |
|00005950| 77 68 69 63 68 20 63 72 | 69 74 65 72 69 6f 6e 2f |which cr|iterion/|
|00005960| 6d 65 74 68 6f 64 20 69 | 73 20 75 73 65 64 20 74 |method i|s used t|
|00005970| 6f 20 73 65 6c 65 63 74 | 20 73 69 67 6e 69 66 69 |o select| signifi|
|00005980| 63 61 6e 74 20 76 61 72 | 69 61 62 6c 65 73 20 61 |cant var|iables a|
|00005990| 6e 64 20 74 68 65 69 72 | 20 70 61 72 61 6d 65 74 |nd their| paramet|
|000059a0| 65 72 73 20 28 77 65 69 | 67 68 74 73 29 20 61 6e |ers (wei|ghts) an|
|000059b0| 64 20 68 6f 77 20 76 61 | 6c 69 64 20 74 68 65 20 |d how va|lid the |
|000059c0| 6d 6f 64 65 6c 20 77 69 | 6c 6c 20 62 65 2e 0d 0d |model wi|ll be...|
|000059d0| 47 65 6e 65 72 61 6c 6c | 79 2c 20 74 68 69 73 20 |Generall|y, this |
|000059e0| 69 73 20 6f 6e 65 20 6f | 66 20 74 68 65 20 6d 6f |is one o|f the mo|
|000059f0| 73 74 20 64 69 66 66 69 | 63 75 6c 74 20 71 75 65 |st diffi|cult que|
|00005a00| 73 74 69 6f 6e 73 20 66 | 6f 72 20 61 6c 6c 20 6d |stions f|or all m|
|00005a10| 65 74 68 6f 64 73 20 77 | 68 69 63 68 20 69 6e 74 |ethods w|hich int|
|00005a20| 65 6e 64 20 74 6f 20 73 | 6f 6c 76 65 20 61 20 62 |end to s|olve a b|
|00005a30| 6c 61 63 6b 2d 62 6f 78 | 20 70 72 6f 62 6c 65 6d |lack-box| problem|
|00005a40| 3a 20 73 74 61 74 69 73 | 74 69 63 73 2c 20 4e 65 |: statis|tics, Ne|
|00005a50| 75 72 61 6c 20 4e 65 74 | 77 6f 72 6b 73 20 28 4e |ural Net|works (N|
|00005a60| 4e 27 73 29 2c 20 47 65 | 6e 65 74 69 63 20 61 6c |N's), Ge|netic al|
|00005a70| 67 6f 72 69 74 68 6d 73 | 2c 20 47 4d 44 48 20 65 |gorithms|, GMDH e|
|00005a80| 74 63 2e 20 54 68 69 73 | 20 69 73 20 70 61 72 74 |tc. This| is part|
|00005a90| 69 63 75 6c 61 72 6c 79 | 20 74 72 75 65 20 66 6f |icularly| true fo|
|00005aa0| 72 20 73 68 6f 72 74 20 | 61 6e 64 20 6e 6f 69 73 |r short |and nois|
|00005ab0| 79 20 64 61 74 61 20 73 | 61 6d 70 6c 65 73 2e 0d |y data s|amples..|
|00005ac0| 0d 49 6e 20 73 74 61 74 | 69 73 74 69 63 73 2c 20 |.In stat|istics, |
|00005ad0| 74 65 73 74 73 20 66 6f | 72 20 73 69 67 6e 69 66 |tests fo|r signif|
|00005ae0| 69 63 61 6e 63 65 20 28 | 74 2d 74 65 73 74 2c 20 |icance (|t-test, |
|00005af0| 46 2d 74 65 73 74 29 20 | 61 72 65 20 75 73 65 64 |F-test) |are used|
|00005b00| 20 74 6f 20 67 65 74 20 | 61 6e 20 69 6e 66 6f 72 | to get |an infor|
|00005b10| 6d 61 74 69 6f 6e 20 6f | 6e 20 74 68 65 20 72 65 |mation o|n the re|
|00005b20| 6c 65 76 61 6e 63 65 20 | 6f 66 20 70 61 72 61 6d |levance |of param|
|00005b30| 65 74 65 72 73 2e 20 48 | 6f 77 65 76 65 72 2c 20 |eters. H|owever, |
|00005b40| 74 68 65 73 65 20 74 65 | 73 74 73 20 68 61 76 65 |these te|sts have|
|00005b50| 20 74 77 6f 20 6d 61 69 | 6e 20 64 69 73 61 64 76 | two mai|n disadv|
|00005b60| 61 6e 74 61 67 65 73 3a | 0d 31 2e 20 54 68 65 20 |antages:|.1. The |
|00005b70| 73 6d 61 6c 6c 65 72 20 | 74 68 65 20 64 61 74 61 |smaller |the data|
|00005b80| 20 73 61 6d 70 6c 65 20 | 61 6e 64 20 74 68 65 20 | sample |and the |
|00005b90| 6d 6f 72 65 20 6e 6f 69 | 73 79 20 69 74 20 69 73 |more noi|sy it is|
|00005ba0| 20 74 68 65 20 6d 6f 72 | 65 20 72 61 6e 64 6f 6d | the mor|e random|
|00005bb0| 6e 65 73 73 20 77 69 6c | 6c 20 62 65 20 63 6f 6d |ness wil|l be com|
|00005bc0| 70 72 65 68 65 6e 64 65 | 64 20 62 79 20 74 68 65 |prehende|d by the|
|00005bd0| 20 6d 6f 64 65 6c 2e 20 | 54 68 65 20 72 65 73 75 | model. |The resu|
|00005be0| 6c 74 73 20 61 72 65 20 | 73 6f 2d 63 61 6c 6c 65 |lts are |so-calle|
|00005bf0| 64 20 6f 76 65 72 66 69 | 74 74 65 64 20 6d 6f 64 |d overfi|tted mod|
|00005c00| 65 6c 73 2e 0d 32 2e 20 | 54 68 65 79 20 70 72 65 |els..2. |They pre|
|00005c10| 73 75 70 70 6f 73 65 20 | 74 68 61 74 20 74 68 65 |suppose |that the|
|00005c20| 20 6e 75 6d 62 65 72 20 | 6f 66 20 70 6f 74 65 6e | number |of poten|
|00005c30| 74 69 61 6c 20 69 6e 70 | 75 74 20 76 61 72 69 61 |tial inp|ut varia|
|00005c40| 62 6c 65 73 20 69 73 20 | 65 71 75 61 6c 20 74 6f |bles is |equal to|
|00005c50| 20 74 68 65 20 6e 75 6d | 62 65 72 20 6f 66 20 76 | the num|ber of v|
|00005c60| 61 72 69 61 62 6c 65 73 | 20 72 65 61 6c 6c 79 20 |ariables| really |
|00005c70| 75 73 65 64 20 69 6e 20 | 74 68 65 20 6d 6f 64 65 |used in |the mode|
|00005c80| 6c 2e 20 55 73 75 61 6c | 6c 79 2c 20 68 6f 77 65 |l. Usual|ly, howe|
|00005c90| 76 65 72 2c 20 6f 6e 65 | 20 69 73 20 69 6e 74 65 |ver, one| is inte|
|00005ca0| 72 65 73 74 65 64 20 69 | 6e 20 66 69 6e 64 69 6e |rested i|n findin|
|00005cb0| 67 20 61 20 73 75 62 73 | 65 74 20 6f 66 20 74 68 |g a subs|et of th|
|00005cc0| 65 20 70 6f 74 65 6e 74 | 69 61 6c 20 69 6e 70 75 |e potent|ial inpu|
|00005cd0| 74 20 76 61 72 69 61 62 | 6c 65 73 20 2d 20 74 68 |t variab|les - th|
|00005ce0| 65 20 6d 6f 73 74 20 72 | 65 6c 65 76 61 6e 74 20 |e most r|elevant |
|00005cf0| 2d 20 66 6f 72 20 6d 6f | 64 65 6c 20 63 6f 6e 73 |- for mo|del cons|
|00005d00| 74 72 75 63 74 69 6f 6e | 20 6f 6e 6c 79 2e 20 44 |truction| only. D|
|00005d10| 75 65 20 74 6f 20 74 68 | 69 73 20 63 6f 6e 74 72 |ue to th|is contr|
|00005d20| 61 64 69 63 74 69 6f 6e | 20 74 68 65 20 72 65 73 |adiction| the res|
|00005d30| 75 6c 74 69 6e 67 20 6d | 6f 64 65 6c 73 20 74 65 |ulting m|odels te|
|00005d40| 6e 64 20 74 6f 20 62 65 | 20 6f 76 65 72 2d 63 6f |nd to be| over-co|
|00005d50| 6d 70 6c 69 63 61 74 65 | 64 20 61 6e 64 20 6d 6f |mplicate|d and mo|
|00005d60| 72 65 20 6f 72 20 6c 65 | 73 73 20 69 6e 73 74 61 |re or le|ss insta|
|00005d70| 62 6c 65 2e 20 41 67 61 | 69 6e 2c 20 74 68 65 79 |ble. Aga|in, they|
|00005d80| 20 77 69 6c 6c 20 6f 76 | 65 72 66 69 74 20 74 68 | will ov|erfit th|
|00005d90| 65 20 64 61 74 61 20 75 | 73 65 64 20 74 6f 20 63 |e data u|sed to c|
|00005da0| 72 65 61 74 65 20 74 68 | 65 20 6d 6f 64 65 6c 20 |reate th|e model |
|00005db0| 28 64 65 73 69 67 6e 20 | 64 61 74 61 29 2e 20 0d |(design |data). .|
|00005dc0| 49 6e 20 6f 74 68 65 72 | 20 77 6f 72 64 73 2c 20 |In other| words, |
|00005dd0| 74 68 65 20 6d 6f 72 65 | 20 70 6f 74 65 6e 74 69 |the more| potenti|
|00005de0| 61 6c 20 69 6e 70 75 74 | 20 76 61 72 69 61 62 6c |al input| variabl|
|00005df0| 65 73 20 61 72 65 20 63 | 6f 6e 73 69 64 65 72 65 |es are c|onsidere|
|00005e00| 64 20 61 20 70 72 69 6f | 72 69 20 66 6f 72 20 6d |d a prio|ri for m|
|00005e10| 6f 64 65 6c 69 6e 67 20 | 28 77 68 69 63 68 20 63 |odeling |(which c|
|00005e20| 61 6e 20 62 65 20 72 61 | 6e 64 6f 6d 20 76 61 6c |an be ra|ndom val|
|00005e30| 75 65 73 20 61 6c 73 6f | 29 20 74 68 65 20 62 65 |ues also|) the be|
|00005e40| 74 74 65 72 20 74 68 65 | 20 6d 6f 64 65 6c 20 22 |tter the| model "|
|00005e50| 71 75 61 6c 69 74 79 22 | 20 77 69 6c 6c 20 62 65 |quality"| will be|
|00005e60| 20 61 63 63 6f 72 64 69 | 6e 67 6c 79 20 74 6f 20 | accordi|ngly to |
|00005e70| 74 68 65 73 65 20 73 69 | 67 6e 69 66 69 63 61 6e |these si|gnifican|
|00005e80| 63 65 20 74 65 73 74 73 | 2e 20 54 68 69 73 20 73 |ce tests|. This s|
|00005e90| 6f 2d 63 61 6c 6c 65 64 | 20 61 72 74 69 66 69 63 |o-called| artific|
|00005ea0| 69 61 6c 20 73 6b 69 6c | 6c 20 69 73 20 64 61 6e |ial skil|l is dan|
|00005eb0| 67 65 72 6f 75 73 20 73 | 69 6e 63 65 20 74 68 65 |gerous s|ince the|
|00005ec0| 20 72 69 73 6b 20 74 68 | 61 74 20 74 68 65 20 6d | risk th|at the m|
|00005ed0| 6f 64 65 6c 20 77 69 6c | 6c 20 72 65 66 6c 65 63 |odel wil|l reflec|
|00005ee0| 74 20 6f 6e 6c 79 20 72 | 61 6e 64 6f 6d 20 72 65 |t only r|andom re|
|00005ef0| 6c 61 74 69 6f 6e 73 68 | 69 70 73 20 69 6e 63 72 |lationsh|ips incr|
|00005f00| 65 61 73 65 73 20 77 69 | 74 68 20 74 68 65 20 6e |eases wi|th the n|
|00005f10| 75 6d 62 65 72 20 6f 66 | 20 70 6f 74 65 6e 74 69 |umber of| potenti|
|00005f20| 61 6c 20 69 6e 70 75 74 | 20 76 61 72 69 61 62 6c |al input| variabl|
|00005f30| 65 73 2e 0d 20 20 20 20 | 20 0d 46 6f 72 20 4e 65 |es.. | .For Ne|
|00005f40| 75 72 61 6c 20 4e 65 74 | 77 6f 72 6b 73 20 74 68 |ural Net|works th|
|00005f50| 65 20 70 72 6f 62 6c 65 | 6d 20 69 73 20 73 69 6d |e proble|m is sim|
|00005f60| 69 6c 61 72 2e 20 48 6f | 77 65 76 65 72 2c 20 4e |ilar. Ho|wever, N|
|00005f70| 4e 27 73 20 74 72 79 20 | 74 6f 20 70 65 72 66 6f |N's try |to perfo|
|00005f80| 72 6d 20 61 20 67 6c 6f | 62 61 6c 20 73 65 61 72 |rm a glo|bal sear|
|00005f90| 63 68 20 6f 6e 20 61 20 | 68 69 67 68 20 64 69 6d |ch on a |high dim|
|00005fa0| 65 6e 73 69 6f 6e 61 6c | 20 61 6e 64 20 6d 75 6c |ensional| and mul|
|00005fb0| 74 69 6d 6f 64 61 6c 20 | 73 75 72 66 61 63 65 20 |timodal |surface |
|00005fc0| 74 6f 20 66 69 6e 64 20 | 61 6e 20 6f 70 74 69 6d |to find |an optim|
|00005fd0| 61 6c 20 6d 6f 64 65 6c | 2e 20 54 68 69 73 20 61 |al model|. This a|
|00005fe0| 70 70 72 6f 61 63 68 20 | 69 73 20 6d 75 63 68 20 |pproach |is much |
|00005ff0| 6d 6f 72 65 20 72 69 73 | 6b 79 2c 20 6d 75 63 68 |more ris|ky, much|
|00006000| 20 6d 6f 72 65 20 73 65 | 6e 73 69 74 69 76 65 20 | more se|nsitive |
|00006010| 61 6e 64 20 6c 65 73 73 | 20 65 66 66 65 63 74 69 |and less| effecti|
|00006020| 76 65 20 69 6e 20 6d 6f | 73 74 20 63 61 73 65 73 |ve in mo|st cases|
|00006030| 20 74 68 61 6e 20 61 64 | 76 61 6e 63 65 64 20 73 | than ad|vanced s|
|00006040| 74 61 74 69 73 74 69 63 | 73 2e 0d 0d 48 6f 77 65 |tatistic|s...Howe|
|00006050| 76 65 72 2c 20 6d 6f 73 | 74 20 69 6d 70 6f 72 74 |ver, mos|t import|
|00006060| 61 6e 74 20 69 73 20 74 | 68 61 74 20 4e 4e 27 73 |ant is t|hat NN's|
|00006070| 20 73 68 6f 77 20 74 68 | 65 20 73 61 6d 65 20 66 | show th|e same f|
|00006080| 61 74 61 6c 20 72 65 6c | 61 74 69 6f 6e 73 68 69 |atal rel|ationshi|
|00006090| 70 20 61 73 20 73 74 61 | 74 69 73 74 69 63 61 6c |p as sta|tistical|
|000060a0| 20 6d 6f 64 65 6c 69 6e | 67 20 6d 65 74 68 6f 64 | modelin|g method|
|000060b0| 73 3a 20 74 68 65 20 6d | 6f 72 65 20 63 6f 6d 70 |s: the m|ore comp|
|000060c0| 6c 65 78 20 74 68 65 20 | 6d 6f 64 65 6c 20 69 73 |lex the |model is|
|000060d0| 20 74 68 65 20 62 65 74 | 74 65 72 20 69 74 73 20 | the bet|ter its |
|000060e0| 22 71 75 61 6c 69 74 79 | 22 20 28 6f 72 2c 20 74 |"quality|" (or, t|
|000060f0| 68 65 20 73 6d 61 6c 6c | 65 72 20 69 74 73 20 65 |he small|er its e|
|00006100| 72 72 6f 72 29 20 77 69 | 6c 6c 20 62 65 2e 20 54 |rror) wi|ll be. T|
|00006110| 68 69 73 20 6d 65 61 6e | 73 2c 20 6f 6e 65 20 63 |his mean|s, one c|
|00006120| 61 6e 20 61 6c 77 61 79 | 73 20 67 65 74 20 61 20 |an alway|s get a |
|00006130| 73 6d 61 6c 6c 20 6d 6f | 64 65 6c 20 65 72 72 6f |small mo|del erro|
|00006140| 72 20 6f 6e 20 64 65 73 | 69 67 6e 20 64 61 74 61 |r on des|ign data|
|00006150| 20 6d 61 6b 69 6e 67 20 | 69 74 20 6d 6f 72 65 20 | making |it more |
|00006160| 63 6f 6d 70 6c 65 78 20 | 28 61 6e 64 20 74 68 75 |complex |(and thu|
|00006170| 73 2c 20 74 6f 20 6f 76 | 65 72 66 69 74 20 74 68 |s, to ov|erfit th|
|00006180| 65 20 6d 6f 64 65 6c 29 | 20 62 79 20 61 64 64 69 |e model)| by addi|
|00006190| 6e 67 20 6d 6f 72 65 20 | 68 69 64 64 65 6e 20 6e |ng more |hidden n|
|000061a0| 65 75 72 6f 6e 73 20 6f | 72 20 61 6e 20 61 64 64 |eurons o|r an add|
|000061b0| 69 74 69 6f 6e 61 6c 20 | 68 69 64 64 65 6e 20 6c |itional |hidden l|
|000061c0| 61 79 65 72 20 74 6f 20 | 74 68 65 20 6e 65 74 77 |ayer to |the netw|
|000061d0| 6f 72 6b 2e 20 54 68 65 | 20 70 65 72 66 6f 72 6d |ork. The| perform|
|000061e0| 61 6e 63 65 20 6f 6e 20 | 6e 65 77 20 64 61 74 61 |ance on |new data|
|000061f0| 2c 20 68 6f 77 65 76 65 | 72 2c 20 6d 69 67 68 74 |, howeve|r, might|
|00006200| 20 62 65 20 63 61 74 61 | 73 74 72 6f 70 68 69 63 | be cata|strophic|
|00006210| 61 6c 20 64 75 65 20 74 | 6f 20 74 68 65 20 6d 6f |al due t|o the mo|
|00006220| 64 65 6c 73 20 72 65 66 | 6c 65 63 74 69 6f 6e 20 |dels ref|lection |
|00006230| 6f 66 20 72 61 6e 64 6f | 6d 6e 65 73 73 2e 20 0d |of rando|mness. .|
|00006240| 0d 54 68 65 20 64 69 6c | 65 6d 6d 61 20 6f 66 20 |.The dil|emma of |
|00006250| 73 74 61 74 69 73 74 69 | 63 73 20 61 6e 64 20 4e |statisti|cs and N|
|00006260| 4e 27 73 20 69 73 20 74 | 68 61 74 20 74 68 65 72 |N's is t|hat ther|
|00006270| 65 20 69 73 20 6e 6f 20 | 72 75 6c 65 20 74 6f 20 |e is no |rule to |
|00006280| 66 69 6e 64 20 61 20 6d | 6f 64 65 6c 20 73 74 72 |find a m|odel str|
|00006290| 75 63 74 75 72 65 2f 63 | 6f 6d 70 6c 65 78 69 74 |ucture/c|omplexit|
|000062a0| 79 20 77 68 69 63 68 20 | 6d 69 67 68 74 20 62 65 |y which |might be|
|000062b0| 20 6f 70 74 69 6d 61 6c | 20 66 6f 72 20 61 20 67 | optimal| for a g|
|000062c0| 69 76 65 6e 20 74 61 73 | 6b 3a 20 69 66 20 61 20 |iven tas|k: if a |
|000062d0| 6d 6f 64 65 6c 20 74 6f | 6f 20 73 69 6d 70 6c 65 |model to|o simple|
|000062e0| 20 69 74 20 70 65 72 66 | 6f 72 6d 73 20 62 61 64 | it perf|orms bad|
|000062f0| 6c 79 2c 20 69 66 20 69 | 74 20 74 6f 6f 20 63 6f |ly, if i|t too co|
|00006300| 6d 70 6c 65 78 20 69 74 | 20 70 65 72 66 6f 72 6d |mplex it| perform|
|00006310| 73 20 62 61 64 6c 79 20 | 6f 6e 20 6e 65 77 20 64 |s badly |on new d|
|00006320| 61 74 61 20 61 6c 73 6f | 2e 20 0d 0d 54 6f 20 61 |ata also|. ..To a|
|00006330| 76 6f 69 64 20 6f 76 65 | 72 66 69 74 74 65 64 20 |void ove|rfitted |
|00006340| 6d 6f 64 65 6c 73 20 61 | 6e 20 61 64 64 69 74 69 |models a|n additi|
|00006350| 6f 6e 61 6c 20 70 72 69 | 6e 63 69 70 6c 65 20 69 |onal pri|nciple i|
|00006360| 73 20 6e 65 65 64 65 64 | 20 77 68 69 63 68 20 6d |s needed| which m|
|00006370| 75 73 74 20 63 6f 6e 73 | 69 64 65 72 20 6d 6f 64 |ust cons|ider mod|
|00006380| 65 6c 20 63 6f 6d 70 6c | 65 78 69 74 79 2e 20 54 |el compl|exity. T|
|00006390| 68 69 73 20 74 61 73 6b | 20 69 73 20 65 71 75 61 |his task| is equa|
|000063a0| 6c 20 74 6f 20 74 68 65 | 20 70 72 6f 62 6c 65 6d |l to the| problem|
|000063b0| 20 6f 66 20 66 69 6e 64 | 69 6e 67 20 61 6e 20 6f | of find|ing an o|
|000063c0| 70 74 69 6d 61 6c 20 6d | 6f 64 65 6c 20 73 74 72 |ptimal m|odel str|
|000063d0| 75 63 74 75 72 65 20 6f | 72 2c 20 69 6e 20 4e 4e |ucture o|r, in NN|
|000063e0| 20 6a 61 72 67 6f 6e 2c | 20 61 6e 20 6f 70 74 69 | jargon,| an opti|
|000063f0| 6d 61 6c 20 6e 65 74 77 | 6f 72 6b 20 74 6f 70 6f |mal netw|ork topo|
+--------+-------------------------+-------------------------+--------+--------+
Only 25.0 KB of data is shown above.
