search

mlandry22 / ieee-fraud-detection

Kaggle ieee-fraud-detection
1 stars 0 forks source link

Archive first model, bedifoca #3

Open mlandry22 opened 4 years ago

mlandry22 commented 4 years ago

image

mlandry22 commented 4 years ago 
Experiment: bedifoca, 2019-08-24 14:08, 1.5.3
      Settings: 6/4/1, seed=267105048, GPUs enabled
      Train data: ieee_train_1.csv (590540, 433)
      Validation data: N/A
      Test data: ieee_test_1.csv (506691, 432)
      Target column: isFraud (binary, 3.499% target class)
System specs: Docker/Linux, 126 GB, 32 CPU cores, 1/1 GPU
      Max memory usage: 93.7 GB, 8.75 GB GPU
Recipe: AutoDL (52 iterations, 4 individuals)
      Validation scheme: stratified, 1 internal holdout
      Feature engineering: 13657 features scored (0 selected)
Timing:
      Data preparation: 150.78 secs
      Model and feature tuning: 3542.80 secs (25 of 32 models trained)
      Feature evolution: 9515.25 secs (112 models trained)
      Final pipeline training: 26965.98 secs (10 models trained)
      Python / MOJO scorer building: 36.72 secs / 0.00 secs
Validation score: AUC = 0.90877 +/- 0.002055 (baseline)
Validation score: AUC = 0.97278 +/- 0.0018174 (final pipeline)
Test score:          AUC = N/A (no target)
mlandry22 commented 4 years ago

still blending with it: image