Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!news.sprintlink.net!redstone.interpath.net!sas!mozart.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: [Q]: Missing Output Values !?
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <D7t1o4.Ctn@unx.sas.com>
Date: Sat, 29 Apr 1995 16:26:28 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References:  <W.ElDeredy.43.000D8FE4@ucl.ac.uk>
Organization: SAS Institute Inc.
Lines: 26


In article <W.ElDeredy.43.000D8FE4@ucl.ac.uk>, W.ElDeredy@ucl.ac.uk (Wael El-Deredy) writes:
|> ...
|> The problem: Some of the training set input vectors don't have their
|> corresponding output vector fully defined. i.e. some outputs are missing (due
|> to a failed or unreliable measurement of that particular response).
|>
|> What would be the best strategy to propagate the error if one (or more) of the
|> target values is (are) missing ?

Missing target values are not a problem.  You just omit those outputs
from the backpropagation computation.  Omitting the entire training case
is pointless. Replacing missing target values with random values or
means or whatever simply confuses the net.

Missing inputs _are_ a problem, for which there are various
unsatisfactory solutions, but that is a different issue entirely from
missing targets.



-- 

Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
(919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.
