Dataset: concrete_compressive_strength (regression)
Penalty: 0.2
Seed: 37
Best fitness: -30.17547304338217
Final val loss: 83.09173376
Final penalty: 0.18886408
Model saved to: logs/regression/concrete_compressive_strength/models/best_model_penalty_0.2_seed_37.pth

Final architecture & hyperparameters:
  num_layers: 4
  layer_sizes: [12, 5, 2, 10]
  activations: [3, 4, 2, 4]
  dropout_rates: [0.019, 0.117, 0.464, 0.021]
  batch_norms: [0, 0, 0, 1]
  learning_rate: 0.0177
  batch_size: 32
  patience: 23
  optimizer_type: 2
  init_type: 1
  l2_penalty: 0.0

Validation metrics (final):
  mae: 7.2748188972473145
  mse: 83.09173583984375
  rmse: 9.115466847059658
  r2_score: 0.6849982738494873
  mape: 36.50538921356201
  residual_std: 9.091843605041504
  prediction_std: 9.015501022338867
  target_std: 16.241363525390625
  normalized_mae: 0.09342261915259338
  normalized_rmse: 0.11706006701736248
  num_samples: 154
  target_range: 77.8699951171875
  prediction_range: 31.300731658935547

Test metrics (final):
  mae: 6.879294395446777
  mse: 75.01492309570312
  rmse: 8.661115580322383
  r2_score: 0.7027032375335693
  mape: 27.19769775867462
  residual_std: 8.649275779724121
  prediction_std: 8.923307418823242
  target_std: 15.884690284729004
  normalized_mae: 0.09412087481362313
  normalized_rmse: 0.1184993297890257
  num_samples: 155
  target_range: 73.08999633789062
  prediction_range: 31.249732971191406
