ensemble_model_score.py

import pandas as pd
import argparse

# Parse the run id from the passed argument, the run id is get from each MLflow experiment run (within the URL)
parser = argparse.ArgumentParser()
parser.add_argument('--run_id', dest='run_id')
args = parser.parse_args()

# generate predict results for each model
for i, model_key in enumerate(model_keys):
  model_key = model_key.strip()
  model_name = f"model_{model_key}"
  model_uri = f"runs:/{args.run_id}/{model_name}"
  model = mlflow.sklearn.load_model(model_uri)
  
  #(load and preprocess data...)
  features = model.features
  X_score = df[features]
  preds = pd.Series(model.predict_proba(X_score)[:, 1])
  preds.name = 'model_score'
  df_pred = pd.concat([df[['model_key']], preds], axis=1)
  df_pred_all = df_pred_all.append(df_pred)
  
# select the best model and result from df_pred_all