14
§
PROC DMNEURL: Approximation to PROC NEURAL
Estimation Stages
6
Max. Number Components
3
Minimum R2 Value
0.000050
Number Grid Points
17
Variable
Mean
Std Dev
Skewness
Kurtosis
LOAN
18608
11207
2.02378
6.93259
MORTDUE
67350
44458
1.81448
6.48187
VALUE
99863
57386
3.05334
24.36280
YOJ
8.15130
7.57398
0.98846
0.37207
DELINQ
0.40570
1.12727
4.02315
23.56545
CLAGE
170.47634
85.81009
1.34341
7.59955
NINQ
1.08456
1.72867
2.62198
9.78651
CLNO
20.50285
10.13893
0.77505
1.15767
DEBTINC
26.59885
8.60175
2.85235
50.50404
For an interval target the percentiles of the response (target) variable are computed
as an aside of the preliminary runs through the data. (Note, that the values of the
response
Ý
are not all stored in RAM.)
Percentiles of Target LOAN in [1100 : 89900]
Nobs
Y Value
Label
1
596
7600.000000
0.073198198
2
1192
10000
0.100225225
3
1788
12100
0.123873874
4
2384
14400
0.149774775
5
2980
16300
0.171171171
6
3576
18800
0.199324324
7
4172
21700
0.231981982
8
4768
25000
0.269144144
9
5364
30500
0.331081081
10
5960
89900
1
The first estimation stage starts with the selection of the best predictor components
(eigenvectors):
Component Selection: SS(y) and R2 (SS_total=326.60303927)
Comp
Eigval
R-Square
F Value
p-Value
SSE
2
14414
0.015964
96.672480
<.0001
321.389163
28
1232.230727
0.005739
34.949673
<.0001
319.514886
11
6335.576701
0.005490
33.620923
<.0001
317.721686
A maximum of 8 iterations is needed for convergence:
Purpose of PROC DMNEURL
§
15
------------ Optimization Cycle (Stage=0) --------------
------------ Activation= SQUARE (Stage=0) --------------
NOTE: ABSGCONV convergence criterion satisfied.
SQUARE: Iter=1 Crit=0.00719589: SSE=6.76374E11 Acc= 32.7484
------------ Activation= TANH (Stage=0) ----------------
NOTE: ABSGCONV convergence criterion satisfied.
TANH: Iter=7 Crit=0.00729363: SSE=6.85561E11 Acc= 29.2423
------------ Activation= ARCTAN (Stage=0) --------------
NOTE: ABSGCONV convergence criterion satisfied.
ARCTAN: Iter=4 Crit=0.00730523: SSE=6.86651E11 Acc= 29.2427
------------ Activation= LOGIST (Stage=0) --------------
NOTE: ABSGCONV convergence criterion satisfied.
LOGIST: Iter=5 Crit=
0.007296: SSE=6.85784E11 Acc= 28.7727
------------ Activation= GAUSS (Stage=0) ---------------
NOTE: ABSGCONV convergence criterion satisfied.
GAUSS: Iter=8 Crit=0.00753243: SSE=7.08006E11 Acc= 15.4180
------------ Activation= SIN (Stage=0) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
SIN: Iter=2 Crit=0.00732518: SSE=6.88526E11 Acc= 29.0399
------------ Activation= COS (Stage=0) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
COS: Iter=5 Crit=0.00753876: SSE=7.08602E11 Acc= 22.5534
------------ Activation= EXP (Stage=0) -----------------
NOTE: ABSGCONV convergence criterion satisfied.
EXP: Iter=2 Crit=0.00724534: SSE=6.81022E11 Acc= 29.9011
For interval target
Ý
the accuracy is computed as the Goodman-Kruskal
coefficient
for a observed-predicted frequency table using the percentiles of
Ý
for row and col-
umn definitions. (Note, that the Goodman-Kruskal
can have negative values for
extrem bad fit.)
Approximate Goodness-of-Fit Criteria (Stage 0)
Run
Activation
Criterion
SSE
Accuracy
1
SQUARE
0.007196
676373814905
32.748384
8
EXP
0.007245
681021814295
29.901149
2
TANH
0.007294
685560807525
29.242251
4
LOGIST
0.007296
685783817673
28.772685
3
ARCTAN
0.007305
686651267193
29.242724
6
SIN
0.007325
688526431720
29.039929
5
GAUSS
0.007532
708006119992
15.418007
7
COS
0.007539
708601735521
22.553358
The Root-Mean-Squared-Estimate RMSE for the first stage is 10589:
Goodness-of-Fit Criteria (Ordered by SSE, Stage 0)