18
§
PROC DMNEURL: Approximation to PROC NEURAL
COS: Iter=4 Crit=0.00671738: SSE=6.31396E11 Acc= 41.2783
------------ Activation= EXP (Stage=2) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
EXP: Iter=0 Crit=0.00656615: SSE=6.17182E11 Acc= 41.6251
The RMSE dropped from 10294 to 10035:
Goodness-of-Fit Criteria (Ordered by SSE, Stage 2)
Run
Activation
SSE
RMSE
Accuracy
4
LOGIST
6.00171E11
10035
44.076324
3
ARCTAN
6.11611E11
10130
42.722853
2
TANH
6.12574E11
10138
42.454925
6
SIN
6.12902E11
10141
42.618536
1
SQUARE
6.14545E11
10154
43.452922
8
EXP
6.17964E11
10183
42.588823
7
COS
6.31415E11
10293
41.153349
5
GAUSS
6.31533E11
10294
41.028769
In stage 3 components are selected w.r.t. the residuals from stage 2:
Component Selection: SS(y) and R2 (Stage=3)
Comp
Eigval
R-Square
F Value p-Value
5
8108.233368
0.008115
48.751302
<.0001
7
7678.598513
0.004569
27.574638
<.0001
27
3496.302840
0.003929
23.802006
<.0001
------------ Optimization Cycle (Stage=3) --------------
------------ Activation= SQUARE (Stage=3) --------------
NOTE: ABSGCONV convergence criterion satisfied.
SQUARE: Iter=1 Crit=0.00627437: SSE=5.89756E11 Acc= 46.5664
------------ Activation= TANH (Stage=3) ----------------
NOTE: ABSGCONV convergence criterion satisfied.
TANH: Iter=2 Crit=0.00628152: SSE=5.90428E11 Acc= 46.2812
------------ Activation= ARCTAN (Stage=3) --------------
NOTE: ABSGCONV convergence criterion satisfied.
ARCTAN: Iter=2 Crit=0.00628195: SSE=5.90469E11 Acc= 46.2125
------------ Activation= LOGIST (Stage=3) --------------
NOTE: ABSGCONV convergence criterion satisfied.
LOGIST: Iter=5 Crit=0.00628127: SSE=5.90404E11 Acc= 45.8040
------------ Activation= GAUSS (Stage=3) ---------------
NOTE: ABSGCONV convergence criterion satisfied.
GAUSS: Iter=2 Crit=0.00637032: SSE=5.98775E11 Acc= 45.3250
------------ Activation= SIN (Stage=3) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
SIN: Iter=6 Crit=0.00627884: SSE=5.90176E11 Acc= 46.0972
Purpose of PROC DMNEURL
§
19
------------ Activation= COS (Stage=3) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
COS: Iter=5 Crit=0.00637738: SSE=5.99438E11 Acc= 44.6439
------------ Activation= EXP (Stage=3) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
EXP: Iter=1 Crit=0.00628498: SSE=5.90753E11 Acc= 46.4650
The RMSE dropped from 10035 to 9939:
Goodness-of-Fit Criteria (Ordered by SSE, Stage 3)
Run
Activation
SSE
RMSE
Accuracy
1
SQUARE
5.88794E11
9939.361833
46.874168
6
SIN
5.89308E11
9943.696786
46.045231
4
LOGIST
5.89532E11
9945.590341
45.936035
2
TANH
5.89658E11
9946.655850
46.521683
3
ARCTAN
5.89714E11
9947.122017
46.462470
8
EXP
5.89994E11
9949.489480
46.590155
5
GAUSS
5.98777E11
10023
45.338954
7
COS
5.99432E11
10029
44.563112
Again, the new stage 4 starts with component selection w.r.t. the residuals of the last
stage3:
Component Selection: SS(y) and R2 (Stage=4)
Comp
Eigval
R-Square
F Value
p-Value
14
5977.581155
0.004044
24.196310
<.0001
24
4589.938565
0.002803
16.817296
<.0001
8
7098.575517
0.002425
14.583721
0.0001
A maximum of 7 iterations is needed for convergence:
------------ Optimization Cycle (Stage=4) --------------
------------ Activation= SQUARE (Stage=4) --------------
NOTE: ABSGCONV convergence criterion satisfied.
SQUARE: Iter=1 Crit=0.00618628: SSE=5.81476E11 Acc= 46.9112
------------ Activation= TANH (Stage=4) ----------------
NOTE: ABSGCONV convergence criterion satisfied.
TANH: Iter=5 Crit= 0.0061812: SSE=5.80998E11 Acc= 45.5545
------------ Activation= ARCTAN (Stage=4) --------------
NOTE: ABSGCONV convergence criterion satisfied.
ARCTAN: Iter=3 Crit=0.00618984: SSE= 5.8181E11 Acc= 45.7876
------------ Activation= LOGIST (Stage=4) --------------
NOTE: ABSGCONV convergence criterion satisfied.
LOGIST: Iter=2 Crit= 0.0062313: SSE=5.85708E11 Acc= 47.2978