ML Pipeline

What this code does

This pipeline trains and evaluates a logistic-regression baseline and records test AUC.

Per experiment, it:

1. Reads outcome data from outcome_folder/outcome_{outcome}.csv.
2. Creates train/val/test split ID files in id_folder/seed{data_seed}/{outcome}/ (if missing).
3. Reads features from feature_folder/feature.csv and aligns them to split IDs.
4. Trains LogisticRegression in a Pipeline(StandardScaler + LogisticRegression) with GridSearchCV (roc_auc).
5. Evaluates on test and appends summary rows to log/auc_summary.csv.

Folder structure expected

• feature_folder/feature.csv
• outcome_folder/outcome_{outcome}.csv
• id_folder/seed{data_seed}/{outcome}/train_ids.csv (auto-created)
• id_folder/seed{data_seed}/{outcome}/val_ids.csv (auto-created)
• id_folder/seed{data_seed}/{outcome}/test_ids.csv (auto-created)

Key files

• create_toy_data.py: generates toy feature.csv and outcome_{outcome}.csv.
• experiments.py: experiment configuration (EXPERIMENTS, sample_id, outcome_col).
• main.py: loops over EXPERIMENTS and writes log/auc_summary.csv.
• run_experiment.py: one experiment run; derives outcome_file as f"outcome_{outcome}.csv".
• save_splits.py: create train, valid, test id split.
• read_data_split.py: reads feature/outcome/split files and builds train/val/test matrices.
• model_pipeline.py: data standardization and model development using cross validation and hyperparameter sesarch
• predict_test.py: use the saved best model to predict test data and evaluate the performance
• evals_plot.py: plot auc curve and confusion matrix

Configure experiments

Edit experiments.py.

Current minimal example:

data_seed = 112
model_seed = 0
sample_id = "sample_id"
outcome_col = "event"

EXPERIMENTS = [
    {
        "outcome": "er",
        "data_seed": data_seed,
        "model_seed": model_seed,
    },
]

Notes:

• outcome drives filename resolution: outcome_folder/outcome_{outcome}.csv.
• sample_id and outcome_col must exist in both feature and outcome files.

Quick start (toy data)

From this directory:

python create_toy_data.py
python main.py

Outputs:

• logs in log/
• AUC summary in log/auc_summary.csv

Input column requirements

• feature.csv must contain:
  • ID column: sample_id (or your configured sample_id)
  • numeric feature columns (all non-ID columns are used as model features)
• outcome_{outcome}.csv must contain:
  • same ID column
  • outcome column named by outcome_col (default event)