This gist was made to demonstrate code on medium.

rf-explanation3.py

 ''' from here we are doing the feature engineering'''
    print(min_leaf_a)
    reset_rf_samples()
    z=RandomForestRegressor(n_jobs=-1,min_samples_leaf= min_leaf_a,max_features= max_feature_a,oob_score=False,n_estimators=40)
    z.fit(X_train,y_train)
    fi=rf_feat_importance(z,df)
    score=0
    final_feature_importance_value=0
    feature_importance_value_list=[0,0.001,0.002,0.0025,0.003,0.0035]
    for feature_importance_value in feature_importance_value_list:
        to_keep = fi[fi.imp>=feature_importance_value].cols
        df_keep = df[to_keep].copy()
        X_train, X_valid = split_vals(df_keep, n_trn)
        feature_tuner=RandomForestRegressor(n_jobs=-1,min_samples_leaf= min_leaf_a,max_features= max_feature_a,oob_score=False,n_estimators=40)
        feature_tuner.fit(X_train,y_train)
        if fx(feature_tuner,X_valid=X_valid,y_valid=y_valid)>=score:
            score=fx(feature_tuner,X_valid=X_valid,y_valid=y_valid)
            final_feature_importance_value=feature_importance_value
    '''Here we will be doing the final model training that is for now, with time i will add more tuning'''
    to_keep = fi[fi.imp>=final_feature_importance_value].cols
    df_keep = df[to_keep].copy()
    X_train, X_valid = split_vals(df_keep, n_trn)
    z=RandomForestRegressor(n_jobs=-1,min_samples_leaf= min_leaf_a,max_features= max_feature_a,oob_score=False,n_estimators=40)
    z.fit(X_train,y_train)

    return[min_leaf_a,max_feature_a,to_keep]