MLProject

name: hp_search

python_env: python_env.yaml

entry_points:
# Use Hyperopt to optimize hyperparams of the train entry_point.
  search_params:
    parameters:
      training_data: {type: string, default: "hemanthsai7/loandefault"}
      max_runs: {type: int, default: 10}
      model_type: {type: str, default: "hgbt"}
    command: "python -O search_params.py {training_data} --max-runs {max_runs} --model-type {model_type}"

  # train Random Forest model with default HPs
  train_rf:
    parameters:
      dset_name: {type: string, default: "sgpjesus/bank-account-fraud-dataset-neurips-2022"}
      max_depth: {type: int, default: 5}
      max_features: {type: float, default: 0.1}
      class_weight: {type: str, default: "balanced"}
      min_samples_leaf: {type: int, default: 10}
    command: "python train_rf.py {dset_name}
                                    --max-depth {max_depth}
                                    --max-features {max_features}
                                    --class-weight {class_weight}
                                    --min-samples-leaf {min_samples_leaf}"

  # train HistGradientBoosted model with default parameters
  train_hgbt:
    parameters:
      dset_name: {type: string, default: "sgpjesus/bank-account-fraud-dataset-neurips-2022"}
      max_depth: {type: int, default: 20}
      learning_rate: {type: float, default: 0.1}
      class_weight: {type: str, default: "balanced"}
      max_leaf_nodes: {type: int, default: 31}
      l2_regularization: {type: int, default: 1.}
    command: "python train_hgbt.py {dset_name}
                                    --max-depth {max_depth}
                                    --learning-rate {learning_rate}
                                    --class-weight {class_weight}
                                    --max-leaf-nodes {max_leaf_nodes}
                                    --l2-regularization {l2_regularization}"