Detail — Mlp

Metrics

• Accuracy: 94.85%
• Macro F1: 94.68%
• Weighted F1: 94.80%
• Best Val Acc: 97.29%

Config (JSON)

{
  "tag": "adamw_swish_bigbatch",
  "layers": [
    640,
    320,
    160
  ],
  "dropouts": [
    0.12,
    0.2,
    0.2
  ],
  "input_dropout": 0.05,
  "activation": "swish",
  "optimizer": "adam",
  "custom_optimizer": "adamw",
  "weight_decay": 0.0001,
  "lr": 0.0008,
  "epochs": 120,
  "batch": 512,
  "use_smote": true,
  "test_size": 0.15,
  "val_size": 0.15,
  "early_stop_patience": 16,
  "reduce_lr": true,
  "rlr_factor": 0.5,
  "rlr_patience": 6
}

Source MD

md/mlp_adamw_swish_bigbatch.md