home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!dtix!darwin.sura.net!mips!sdd.hp.com!ux1.cso.uiuc.edu!news.cso.uiuc.edu!uxa.cso.uiuc.edu!ast48365
- From: ast48365@uxa.cso.uiuc.edu (A Tull )
- Newsgroups: comp.ai.neural-nets
- Subject: beginner's question in neural net...
- Summary: choice of appropriate data
- Keywords: testing data vs training data
- Message-ID: <BrvDrA.7JG@news.cso.uiuc.edu>
- Date: 24 Jul 92 01:35:33 GMT
- Sender: usenet@news.cso.uiuc.edu (Net Noise owner)
- Organization: University of Illinois at Urbana
- Lines: 29
-
- I was trying to implement a net for a simple mathematical function. My first
- question is : should the TRAINING data be deterministic or random ? I was
- trying to make a training set of about two hundred training pairs, but my
- net failed to converge. Is it alright to split my training set into, say,
- 3 smaller training set, and re-train the net ? That is, I feed in the first
- set of data, (assuming convergence), RESET the net, feed in the seconde
- set of data, RESET, feed in the third set of data... and then, gather all
- 3 outputs , and put together to observe how the net learned, will the
- final result make any sense ? (or do I have to analyse them each one on
- their own ?
-
- Another case is, assume now I have 3 training sets and 3 testing sets,
- will the followsing 2 sequence make any difference (or sense) to the
- final output ?
-
- train1, test1, train2, test2, train3, test3
- -OR-
- train1,train2,train3,test1,test2,test3
-
-
- help DESPARATELY in need.
- thanx in advance.
-
- Alice Tull
- tull@ibma0.cs.uiuc.edu
-
- --
- Alice Tull
- ast48365@uxa.cso.uiuc.edu
-