Model #1

Parameters

• training dataset size: 235,738
• testing dataset size: 28,124
• hidden units: [100, 75, 50, 25]

Results

• accuracy: 0.610902
• accuracy_baseline: 0.5
• auc: 0.657483
• auc_precision_recall: 0.644627
• average_loss: 0.655231
• global_step: 235882
• label/mean: 0.5
• loss: 26.1943
• prediction/mean: 0.524407

Model #2

Parameters

• training dataset size: 800,000
• testing dataset size: 88,888
• hidden units: [23, 13, 3]

Results

• accuracy: 0.583858
• accuracy_baseline: 0.5
• auc: 0.619257
• auc_precision_recall: 0.611224
• average_loss: 0.673288
• global_step: 21142445
• label/mean: 0.5
• loss: 13.4655
• prediction/mean: 0.499873

Model #3

Using just my own replays.

Parameters

• total replays: 782
  • training dataset size: 1,404
  • testing dataset size: 160
• hidden units: [23, 13, 3]

Results

• accuracy: 0.5875
• accuracy_baseline: 0.5
• auc: 0.589453
• auc_precision_recall: 0.625588
• average_loss: 0.722218
• global_step: 589682
• label/mean: 0.5
• loss: 1.44444
• prediction/mean: 0.486236