20
§
PROC DMNEURL: Approximation to PROC NEURAL
------------ Activation= GAUSS (Stage=4) ---------------
NOTE: ABSGCONV convergence criterion satisfied.
GAUSS: Iter=3 Crit=0.00617962: SSE= 5.8085E11 Acc= 46.3454
------------ Activation= SIN (Stage=4) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
SIN: Iter=7 Crit=0.00614634: SSE=5.77722E11 Acc= 45.3470
------------ Activation= COS (Stage=4) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
COS: Iter=2 Crit=0.00619158: SSE=5.81974E11 Acc= 47.5278
------------ Activation= EXP (Stage=4) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
EXP: Iter=2 Crit=0.00620074: SSE=5.82835E11 Acc= 47.1760
The RMSE dropped from 9939 to 9889:
Goodness-of-Fit Criteria (Ordered by SSE, Stage 4)
Run
Activation
SSE
RMSE
Accuracy
5
GAUSS
5.82844E11
9889.013804
46.529424
1
SQUARE
5.82906E11
9889.541858
47.202054
7
COS
5.83336E11
9893.184747
47.340247
8
EXP
5.83553E11
9895.031798
47.907199
2
TANH
5.8489E11
9906.353245
46.238301
4
LOGIST
5.86142E11
9916.953006
47.605286
3
ARCTAN
5.88716E11
9938.707816
46.013181
6
SIN
6.12035E11
10134
45.185468
For space reasons we are skipping the results of stage 5 except the following table
which shows that the RMSE dropped again.
Goodness-of-Fit Criteria (Ordered by SSE, Stage 5)
Run
Activation
SSE
RMSE
Accuracy
1
SQUARE
5.78114E11
9848.803178
47.337721
8
EXP
5.78394E11
9851.192061
47.342324
4
LOGIST
5.78507E11
9852.150383
47.269327
3
ARCTAN
5.79057E11
9856.832196
46.609720
2
TANH
5.8133E11
9876.166691
46.144529
5
GAUSS
5.82144E11
9883.077740
46.792103
7
COS
5.82405E11
9885.287904
46.540469
6
SIN
6.10243E11
10119
45.311873
This is a summary table for the first six estimation stages:
Summary Table Across Stages
Stage Activation Link
SSE
RMSE Accuracy
AIC
Purpose of PROC DMNEURL
§
21
0 SQUARE
IDENT 6.68237E11
10589 33.92584 110675
1 SQUARE
IDENT 6.31521E11
10294 41.27589 110544
2 LOGIST
IDENT 6.00171E11
10035 44.07632 110447
3 SQUARE
IDENT 5.88794E11 9939.36183 46.87417 110539
4 GAUSS
IDENT 5.82844E11 9889.01380 46.52942 110684
5 SQUARE
IDENT 5.78114E11 9848.80318 47.33772 110842
The six stages took 48 optimizations (each with 7 parameters) and 33 runs through
the data. In average less than 4 iterations and about 7 function calls are needed for
each optimization:
*** Total Number of Runs through Data :
33
*** Total Number of NL Optimizations
:
48
*** Total Number of Iterations in NLP :
159
*** Total Number Function Calls in NLP:
348
Missing Values
Observations with missing values in the target variable (response or dependend vari-
able) are not included in the analysis. Those observations are, however, scored, i.e.
predicted values are computed.
Observations with missing values in the predictor variables (independend variables)
are processed depending on the scale type of the variable:
¯
For numeric variables, missing values are replaced by the (weighted) mean of
the variable.
¯
For class variables, missing values are treated as an additional category.
Syntax of PROC DMNEURL
ÈÊ
Ç
ÅÆ
ÍÊÄ
ÓÔØ
ÓÒ×
Ö
ÕÙ
Ö
×Ø
Ø
Ñ
Ò
Ø
Î
Ê
Ú
Ö
Ð
×
Ì
Ê
Ì
Ú
Ö
Ð
×
ÍÆ
ÌÁÇÆ
Ò
Ñ
×
ÄÁÆÃ
Ò
Ñ
Ê
É
Ú
Ö
Ð
×
Ï
Á
ÀÌ
Ú
Ö
Ð
×
ÁËÁÇÆ
ÓÔØ
ÓÒ×
ÓÔØ
ÓÒ
Ð
×Ø
Ø
Ñ
Ò
Ø×
Overview of PROC DMNEURL Options
PROC DMNEURL options ;
This statement invokes the DMNEURL procedure. The options available with the
PROC DMNEURL statement are: