>>>>>>>> SAS Notes -- Copyright (c) 1991 - 2000 by SAS Institute Inc. <<<<<<<<
V6-NEURAL-G772 *** ALERT NOTE *** PRODUCT: ENT MINER
PROCUDURE:NEURAL
TITLE: Enterprise Miner Neural Network Fit Statistics May Be Incorrect
KEYS: NEURAL NETWORK FIT STATISTICS INCORRECT RESULTS DENOMINATOR SAMPLE
AVERAGE PROFIT ERROR DF DEGREES OF FREEDOM AIC SBC INCORROUT
The Neural Network Node is not using observations with missing values
for any of the input variables in the computation of fit statistics,
as other nodes are. Therefore, the fit statistics from neural network
models SHOULD NOT be compared with the fit statistics from other
modeling nodes, if there are missing values for any of the input
variables. Also, neural network models should not be compared with
other neural network models based on different sets of inputs with
different patterns of missing values. Any such comparisons will produce
erroneous results, since the nodes are essentially analyzing different
data sets.
If there are no missing values for any of the variables used in the
analysis, comparing the fit statistics from neural network models with
models from other modeling nodes is valid. Also, comparisons between
different neural network models are valid, even with missing inputs, if
they contain the same input (and target) variables.
+------REPORTED------+ +---FIXED---+
SYSTEM RELEASE LEVEL RELEASE LEVEL
WIN/NT 2.0
Win95 2.0
AIX/R 2.01
DigitUnx 2.01
HP800 2.01
Solaris 2.01
WIN/NT 2.01
Win95 2.01
AIX/R 2.02
DigitUnx 2.02
HP800 2.02
Solaris 2.02
WIN/NT 2.02
Win95 2.02
AIX/R 2.03
DigitUnx 2.03
HP800 2.03
Solaris 2.03
WIN/NT 2.03
Win95 2.03
AIX/R 3.0
DigitUnx 3.0
HP800 3.0
Solaris 3.0
WIN/NT 3.0
Win98 3.0
Win95 3.0
AIX/R 3.01
DigitUnx 3.01
HP800 3.01
Solaris 3.01
WIN/NT 3.01
Win98 3.01
Win95 3.01
This archive was generated by hypermail 2b29 : Mon Jun 12 2000 - 14:40:33 EDT