16
§
PROC DMNEURL: Approximation to PROC NEURAL
Run
Activation
SSE
RMSE
Accuracy
1
SQUARE
6.68237E11
10589
33.925841
2
TANH
6.74156E11
10635
30.776431
4
LOGIST
6.74885E11
10641
30.492000
8
EXP
6.77111E11
10659
30.503925
3
ARCTAN
6.78503E11
10670
29.832932
6
SIN
6.81758E11
10695
29.402115
5
GAUSS
7.02918E11
10860
21.174016
7
COS
7.04934E11
10876
20.022492
The second stage (stage=1) starts with selecting the best principal components for
predicting the residual:
Component Selection: SS(y) and R2 (Stage=1)
Comp
Eigval
R-Square
F Value
p-Value
1
16197
0.023135
141.126005
<.0001
3
12170
0.017130
106.340108
<.0001
20
5623.081574
0.012121
76.193667
<.0001
Now a maximum of 5 iterations is needed for convergence:
------------ Optimization Cycle (Stage=1) --------------
------------ Activation= SQUARE (Stage=1) --------------
NOTE: ABSGCONV convergence criterion satisfied.
SQUARE: Iter=1 Crit=0.00675824: SSE=6.35237E11 Acc= 39.9782
------------ Activation= TANH (Stage=1) ----------------
NOTE: ABSGCONV convergence criterion satisfied.
TANH: Iter=4 Crit=0.00677155: SSE=6.36489E11 Acc= 40.1296
------------ Activation= ARCTAN (Stage=1) --------------
NOTE: ABSGCONV convergence criterion satisfied.
ARCTAN: Iter=4 Crit=0.00675928: SSE=6.35335E11 Acc= 41.0832
------------ Activation= LOGIST (Stage=1) --------------
NOTE: ABSGCONV convergence criterion satisfied.
LOGIST: Iter=5 Crit=0.00676491: SSE=6.35864E11 Acc= 41.1768
------------ Activation= GAUSS (Stage=1) ---------------
NOTE: ABSGCONV convergence criterion satisfied.
GAUSS: Iter=5 Crit=0.00701114: SSE=6.59009E11 Acc= 37.5715
------------ Activation= SIN (Stage=1) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
SIN: Iter=4 Crit=0.00676857: SSE=6.36208E11 Acc= 40.1257
------------ Activation= COS (Stage=1) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
COS: Iter=5 Crit=0.00699351: SSE=6.57351E11 Acc= 36.1925
------------ Activation= EXP (Stage=1) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
EXP: Iter=1 Crit=0.00676817: SSE= 6.3617E11 Acc= 40.3710
Purpose of PROC DMNEURL
§
17
The RMSE dropped from 10589 to 10294:
Goodness-of-Fit Criteria (Ordered by SSE, Stage 1)
Run
Activation
SSE
RMSE
Accuracy
1
SQUARE
6.31521E11
10294
41.275893
3
ARCTAN
6.32669E11
10303
40.218692
8
EXP
6.32795E11
10304
40.947613
6
SIN
6.32908E11
10305
40.499603
2
TANH
6.3331E11
10308
40.464591
4
LOGIST
6.33355E11
10309
40.566497
7
COS
6.56699E11
10497
36.381117
5
GAUSS
6.58666E11
10513
37.785522
The third stage starts with selecting the best eigenvectors for prediction of the resid-
uals of the last stage:
Component Selection: SS(y) and R2 (Stage=2)
Comp
Eigval
R-Square
F Value
p-Value
23
4811.081772
0.011805
71.186100
<.0001
28
1232.230727
0.007479
45.434565
<.0001
18
5865.674624
0.006724
41.123895
<.0001
Now, the maximum of iterations is four!
------------ Optimization Cycle (Stage=2) --------------
------------ Activation= SQUARE (Stage=2) --------------
NOTE: ABSGCONV convergence criterion satisfied.
SQUARE: Iter=1 Crit=0.00652258: SSE=6.13086E11 Acc= 42.4729
------------ Activation= TANH (Stage=2) ----------------
NOTE: ABSGCONV convergence criterion satisfied.
TANH: Iter=3 Crit=0.00650972: SSE=6.11878E11 Acc= 41.5654
------------ Activation= ARCTAN (Stage=2) --------------
NOTE: ABSGCONV convergence criterion satisfied.
ARCTAN: Iter=3 Crit=0.00650403: SSE=6.11342E11 Acc= 42.2974
------------ Activation= LOGIST (Stage=2) --------------
NOTE: ABSGCONV convergence criterion satisfied.
LOGIST: Iter=4 Crit=0.00641908: SSE=6.03358E11 Acc= 43.3779
------------ Activation= GAUSS (Stage=2) ---------------
NOTE: ABSGCONV convergence criterion satisfied.
GAUSS: Iter=2 Crit=0.00671802: SSE=6.31456E11 Acc= 41.2660
------------ Activation= SIN (Stage=2) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
SIN: Iter=4 Crit=0.00651654: SSE=6.12519E11 Acc= 41.9809
------------ Activation= COS (Stage=2) -----------------
NOTE: ABSGCONV convergence criterion satisfied.