home *** CD-ROM | disk | FTP | other *** search
- The defining characteristics (above the dashed line) and classifying
- characteristics (below the dashed line) must be defined before anything
- else can be accomplished. If using the analog version, the minimum and
- maximum values are also set here. After defining, you can later use
- this process to change the characteristics. Don't leave any blank
- entries except at the bottom (those will be removed for you).
- *********************** Keys available *****************************
- Esc - return to the Main Menu.
- Up/down arrows, Page Up/Page Down - scroll over the entries.
- Enter - use it to enter data entry mode. Use it again to end edit of an
- entry. Esc will terminate edit mode.
- Insert/Delete - insert an entry before the menu selection bar, or delete
- the entry under the bar. In data entry, Insert toggles insertion
- mode for characters and Delete deletes characters.
- F2 - clear entry when in data entry mode.
- F4/F5 - copy(F4) or paste(F5) entry under the menu selection bar.
- F9 - display fields in file (database version only).
- Home/End - in data entry, go to start/end of string.
- #
- Sample cases have to be entered in order to be learned. In the binary
- version, the Enter key will "turn on" a characteristic by highlighting.
- Be sure to adjust monitor controls to see the highlighting. In the analog
- version, values must be entered in data entry mode (strike Enter). The
- last case may not be blank since blank cases at the end are removed.
- Make sure there are enough sample cases to be comprehensive and represent
- all aspects of the problem.
- *********************** Keys available *****************************
- Esc - return to the Main Menu.
- Up/down arrows, Page Up/Page Down - scroll over the entries.
- Left/Right arrows - scroll from case to case.
- Enter - use it to enter data entry mode (analog version) or highlight a
- characteristic (binary version). End data entry with Esc.
- Insert/Delete - insert/delete the entire current sample case. In data
- entry, Insert toggles insertion mode, Delete deletes characters.
- F2 - clear all characteristics of the current sample case.
- F4/F5 - copy(F4) or paste(F5) an entire sample case.
- F6 - goto a specific case. Requests specific case number.
- Home/End - goto first/last sample case. In data entry, start/end string.
- F7 - Toggles a value to its maximum/minimum in analog version.
- F8 - Toggles bar graphing of characteristic values on and off.
- #
- Before learning, check the sample cases for duplicates with this menu
- entry. The case numbers of all cases whose defining characteristics
- match are shown, whether or not the classifying characteristics match.
- You must decide if it is appropriate to have these duplicates.
- If the sample cases were obtained from an expert and two or more that
- match have different classifications, then only one must be correct.
- If cases were obtained from actual data and two or more that match
- have different classifications, then you will have to decide if this is
- natural or if there are errors.
- If the problem characteristics file is oversize, only the segment
- currently loaded into memory can be checked.
- *********************** Keys available *****************************
- Esc - return to the Main Menu.
- Enter - view the next screen of duplicates if there are more than will
- fit on a screen. When all screens have been viewed, the status and
- data entry bar states "Match checking complete".
- #
- After sample cases have been entered, NeuroShell can learn them with
- this menu entry. Learning stops when the error factors for all sample
- cases fall below the learning threshold. You can stop learning at any
- time with the Esc key, and resume later because the contents of the net-
- work are saved automatically on disk. Speed of learning is affected by
- use of a math coprocessor and these factors:
- number of characteristics, number of hidden nodes, number of sample
- cases, learning threshold, learning rate, momentum, and presentation.
- For more details, see the manual or help for "Set network factors".
- As learning proceeds, the count of cases learned (learning events) is
- displayed in the status and data entry bar. This count is NOT a count of
- learning sweeps through all sample cases.
- The Summary of Learning Error Factors window is usually updated every
- 50 learning events. See Help screen for "Display Summary of Learning
- Error Factors" for more details.
- *********************** Keys available *****************************
- Esc - interrupts learning at the end of the next learning event. It may
- take a few seconds.
- #
- After learning, NeuroShell is ready to classify new cases. New cases
- are entered like sample cases are entered. Unlike sample cases, new cases
- are not automatically adjusted if you later insert or delete character-
- istic definitions. After characteristics for a new case are set, classify
- the case with the F3 key. Later, as you scroll from case to case, each
- case will be classified when you come to it.
- *********************** Keys available *****************************
- Esc - return to the Main Menu.
- Up/down arrows, Page Up/Page Down - scroll over the entries.
- Left/Right arrows - scroll from case to case.
- Enter - use it to enter data entry mode (analog version) or highlight a
- characteristic (binary version). End data entry with Esc.
- Insert/Delete - insert/delete the entire current new case. In data
- entry, Insert toggles insertion mode, Delete deletes characters.
- F2 - clear all characteristics of the current case.
- F3,F9 - F3 classifies the current case. F9 classifies all cases.
- F4/F5 - copy(F4) or paste(F5) an entire case.
- F6 - goto a specific case. Requests specific case number.
- Home/End - go to first/last case. In data entry, start/end of string.
- F7 - Toggles a value to its maximum/minimum in analog version.
- F8 - Toggles bar graphing of characteristic values on and off.
- #
- This function prints the definitions of the characteristics on the
- printer. The analog version prints minimum and maximum values as well.
- Make sure the printer is turned on, in ready status (on line), and
- connected to the port specified when you ran the install program. The
- port is easily changed by selecting "Set hardware system factors" from
- the Advanced Options menu. If there is an error or DOS cannot access the
- printer, it will take DOS a few seconds before the error is sent to
- NeuroShell.
- *********************** Keys available *****************************
- Esc - interrupts printing. Print lines already sent to the printer will
- continue to print until finished.
- #
- This function prints the sample cases on the printer. The analog
- version prints each characteristic with its value, but the binary version
- prints only true (highlighted) characteristics.
- Make sure the printer is turned on, in ready status (on line), and
- connected to the port specified when you ran the install program. The
- port is easily changed by selecting "Set hardware system factors" from
- the Advanced Options menu. If there is an error or DOS cannot access the
- printer, it will take DOS a few seconds before the error is sent to
- NeuroShell.
- *********************** Keys available *****************************
- Esc - interrupts printing. Print lines already sent to the printer will
- continue to print until finished.
- #
- This function prints the new cases to be classified. The analog
- version prints each characteristic with its value, but the binary version
- prints only true (highlighted) characteristics (values of the output
- nodes which are greater than the output threshold are printed also).
- Make sure the printer is turned on, in ready status (on line), and
- connected to the port specified when you ran the install program. The
- port is easily changed by selecting "Set hardware system factors" from
- the Advanced Options menu. If there is an error or DOS cannot access the
- printer, it will take DOS a few seconds before the error is sent to
- NeuroShell.
- *********************** Keys available *****************************
- Esc - interrupts printing. Print lines already sent to the printer will
- continue to print until finished.
- #
- Choosing "Select from Advanced Options menu" makes available a second
- menu with slightly more involved options.
- *********************** Keys available *****************************
- Esc - return to the Main Menu.
- Up/down arrows, Page Up/Page Down - scroll over the selections.
- Enter - use it to select the menu function you want.
- #
- This selection exits from NeuroShell and returns you to DOS. All your
- definitions, sample cases, cases to be classified, and network contents
- have been saved automatically already (unless you received an error
- message alerting you to a problem).
- #
- Use this selection to load in a new problem. You will be prompted for
- the new problem name. Enter a binary problem name if you are in the
- binary version, or an analog problem name if in the analog version. All
- data for the problem you were working on (definitions, cases, network)
- has already been saved (unless you received a message saying there was
- an error). Remember, all problem files must be on the NeuroShell disk or
- directory. The name cannot be over 8 characters long.
- At the prompt for a problem name, you may strike Enter to see a menu
- of all existing problems in the current directory. To select one, scroll
- to it and strike Enter again.
- If you select a problem that doesn't match the version you are using,
- e.g. binary problem when you are in analog, the resulting error message
- has an option that will load the correct version for you.
- *********************** Keys available *****************************
- Esc - abort entry of the problem name and return to the Main Menu.
- Enter - Terminate the entry of the problem name.
- Insert/Delete - Insert toggles insertion mode for characters and Delete
- deletes characters.
- F2 - clear any characters already typed in.
- Home/End - go to start/end of string.
- #
- This function reinitializes the network, thus forgetting all previous
- learning. Use it if you want to relearn with a new learning rate,
- presentation, or momentum. (If you change the number of hidden nodes or
- characteristics, NeuroShell automatically forgets for you because the
- network is then no good.) Forgetting doesn't change the .NET file on
- disk, however, until the next time you learn. So before then, you could
- change your mind about forgetting by loading the problem again.
- *********************** Keys available *****************************
- none
- #
- The error factors summary window shows the number of factors in each
- of several ranges. Each sample case has a learning error factor which is
- based on the differences between what the network is currently computing
- for classifying characteristics and the correct classifying character-
- istics. Learning terminates when all of these factors are below the
- learning threshold set by you (see "Set network factors").
- The column labeled "count" is the number of cases which have factors
- in the range to the left. The percentage this count is of the total is
- shown in the column to the right of the count column.
- If the problem characteristics file is oversize, this window only
- displays the error factors for the last segment learned.
- To see the actual learning error factors, select the next menu entry,
- "Display Learning Error Factors".
- *********************** Keys available *****************************
- Esc - return to the Advanced Options Menu.
- #
- The display of learning error factors shows the actual error factor
- for each sample case. Each sample case has a learning error factor which
- is based on the differences between what the network is now computing
- for classifying characteristics and the correct classifying character-
- istics. The error factors are in order by sample case number from left to
- right, top to bottom, in 20 rows of 9 columns each. If there are more
- than 180 sample cases, strike Enter to see the next group of 180 factors.
- If during learning you notice from the summary that a few of the error
- factors haven't become much smaller after considerable learning, use this
- selection to view the actual factors. Match the high error factors with
- respective sample cases. Then check those sample cases to see if they are
- classified properly or conflict with each other or other cases. If you
- find errors in the sample cases, correct them, forget, and relearn.
- If the problem characteristics file is oversize, this window only
- displays the error factors for the last segment learned.
- Error factors are arbitrarily initialized to 3.
- *********************** Keys available *****************************
- Esc - return to the Advanced Options Menu.
- Enter - view next screen of error factors if there are more than 180.
- #
- This selection displays the amount of memory free in bytes and "k". It
- also displays the name of the problem currently loaded.
- The amount of memory available does not include all small fragments of
- memory that were used and released, so the amount is not 100% accurate.
- NeuroShell will not allow you to completely fill available memory by
- changing network factors, because it reserves a few thousand bytes for
- internal program functions.
- *********************** Keys available *****************************
- none
- #
- A number of network factors can be changed from their defaults if
- necessary. When you change one of these factors, the .FIG file is written
- to disk again to save the new values with the problem. The legal range
- of values for each factor is displayed to the right in parentheses.
- Output threshold - highlight classifying characteristics above this.
- Learning threshold - when error factors go below it learning stops.
- Hidden nodes - default = 2*sqrt(defining chars. + classifying chars.)
- Learning rate - .4 to .6 is best, sometimes .1 or .2 works better.
- Momentum - try lowering it as you lower learning rate.
- Maximum cases in memory - max sample cases loaded in memory at once.
- Maximum characteristics definitions - includes the line of dashes.
- Presentation - whether sample cases are learned randomly or not.
- Characteristics: digits left - number of Xs on screen: XXXXXX.000000.
- Characteristics: digits right - number of Xs on screen: 000000.XXXXXX.
- *********************** Keys available *****************************
- Esc - return to Main Menu after storing values; also aborts an edit.
- Up/down arrows, Page Up/Page Down - scroll over the entries.
- Enter - select value to edit and use again when editing complete.
- Insert/Delete/F2/Home/End - used only in data entry mode as usual.
- #
- This selection performs the same function as the Install program,
- telling NeuroShell about your hardware characteristics. There are three
- categories and only one entry in each category may be selected. Select
- by striking the Enter key, and NeuroShell will highlight the selection.
- The first category defines how NeuroShell interfaces with the monitor.
- True IBM compatible processors work fastest when NeuroShell can write
- directly to the monitor's RAM. If that doesn't work, use BIOS calls.
- The second category of 5 choices selects the type of adaptor. Some
- monitors have only 2 colors like monochrome monitors, but accept color
- interface and sometimes show shades of gray. Choose black and white for
- these, which are often found on portables. If you have a black and white
- where the characters are the dark color, highlighting may appear less
- bold than unhighlighted characters. Try reverse highlight for those. If
- you can't get highlighting any other way, use reverse video.
- The last category tells NeuroShell where to find the printer. Most
- printers are on the first parallel port.
- *********************** Keys available *****************************
- Esc - return to the Main Menu after storing new values.
- Up/down arrows, Page Up/Page Down - scroll over the entries.
- Enter - select an entry to be highlighted.
- #
- When you make this menu selection, you will begin running DOS,
- where you can execute DOS commands. You may want to use this facility
- to erase or load files, or to call your favorite editor, etc. NeuroShell
- remains loaded and its execution will resume when you give DOS the EXIT
- command.
- #
- This selection allows you to select another disk or directory from
- which NeuroShell will read and write files. You will be prompted for
- the DOS path name, and the current path will already be displayed. You
- can erase the current path (F2) or modify it.
- *********************** Keys available *****************************
- Esc - terminate operation without changing the path.
- Enter - change to the path just entered.
- Insert/Delete - insert toggles insertion mode for characters and Delete
- deletes characters.
- F2 - clear entry.
- Home/End - go to start/end of string.
- #
- This selection will enable you to call another program while keeping
- the current NeuroShell session active. You might want to call a program
- that loads data into the .CHR or .CLS file, for example. When the called
- program terminates, it will return to NeuroShell. An error will occur if
- the program is not found or there is not enough memory to load it. You
- will be asked to type in the program name, and the current path will
- already be displayed for your information. You can use this path to
- precede the program name, or you may change the path. If no path is
- specified, the current directory is searched anyway.
- *********************** Keys available *****************************
- Esc - terminate operation without calling a program.
- Enter - call the program just entered.
- Insert/Delete - insert toggles insertion mode for characters and Delete
- deletes characters.
- F2 - clear entry.
- Home/End - go to start/end of string.
- #
- This selection exits from NeuroShell, then loads and executes the
- NeuroShell Utility program. The name of the problem currently in
- NeuroShell is passed to and loaded by the Utility.
- The NeuroShell Utility can graph data (a supported graphics monitor
- is required), import and export files, display and print connection
- weights and error factors, set minimum and maximum values for analog
- problems, randomly extract sample cases to build a classification file
- (test set), and other useful operations.
-
-