Chess Position Evaluation Using Radial Basis Function Neural Networks
Table 3
Aggregate presentation of indicators with mating evaluation filtering.
Scenario
Method
MAE
R2
Training time (s)
Nodes (fuzzy sets)
Testing
Validation
Testing
Validation
1 (3 low-depth evaluation inputs)
RBF
0.76
0.76
0.50
0.52
7304
38219 (25)
CNN
0.94 (1.01 ± 0.06)
0.93 (1.01 ± 0.06)
0.31 (0.30 ± 0.01)
0.32 (0.31 ± 0.01)
3563
—
MLP
0.75 (0.82 ± 0.04)
0.75 (0.81 ± 0.04)
0.46 (0.42 ± 0.02)
0.52 (0.48 ± 0.03)
642
[20 10] (−)
MLP-bitmap inputs
0.70 (0.78 ± 0.06)
0.71 (0.79 ± 0.06)
0.50 (0.47 ± 0.02)
0.50 (0.48 ± 0.02)
9480
[20 10] (−)
2 (2 low-depth evaluation inputs)
RBF
0.83
0.84
0.43
0.45
12448
38229 (21)
CNN
0.98 (1.05 ± 0.07)
0.98 (1.04 ± 0.07)
0.25 (0.25 ± 0.007)
0.26 (0.25 ± 0.008)
3624
—
MLP
0.93 (1.00 ± 0.06)
0.93 (1.00 ± 0.06)
0.36 (0.30 ± 0.04)
0.38 (0.33 ± 0.03)
623
[20 10] (−)
MLP-bitmap inputs
0.85 (1.04 ± 0.17)
0.85 (1.04 ± 0.17)
0.40 (0.35 ± 0.06)
0.42 (0.37 ± 0.04)
9342
[20 10] (−)
3 (no low-depth evaluation inputs)
RBF
0.85
0.85
0.41
0.43
5745
38238 (21)
CNN
1.01 (1.06 ± 0.06)
1.00 (1.06 ± 0.06)
0.23 (0.20 ± 0.007)
0.25 (0.24 ± 0.008)
3902
—
MLP
0.96 (1.05 ± 0.04)
0.97 (1.04 ± 0.04)
0.32 (0.26 ± 0.03)
0.34 (0.30 ± 0.03)
598
[20 10] (−)
MLP-bitmap inputs
0.94 (1.02 ± 0.06)
0.95 (1.02 ± 0.06)
0.35 (0.28 ± 0.05)
0.35 (0.29 ± 0.04)
9211
[20 10] (−)
The table depicts the best performance in the respective dataset in terms of MAE along with the mean value and standard deviation in parenthesis wherever applicable. The best result in terms of MAE in each scenario is marked with bold text.
