Entrenamiento de modelos en sklearn con GridSearchCV y MLflow

• Ultima modificación: Mayo 14, 2022

Carga de datos

def load_data():

    import pandas as pd

    url = "http://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv"
    df = pd.read_csv(url, sep=";")

    y = df["quality"]
    x = df.copy()
    x.pop("quality")

    return x, y

Particionamiento de los datos

def make_train_test_split(x, y):

    from sklearn.model_selection import train_test_split

    (x_train, x_test, y_train, y_test) = train_test_split(
        x,
        y,
        test_size=0.25,
        random_state=123456,
    )
    return x_train, x_test, y_train, y_test

Cálculo de métricas de evaluación

def eval_metrics(y_true, y_pred):

    from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score

    mse = mean_squared_error(y_true, y_pred)
    mae = mean_absolute_error(y_true, y_pred)
    r2 = r2_score(y_true, y_pred)

    return mse, mae, r2

Reporte de métricas de evaluación

def report(estimator, mse, mae, r2):

    print(estimator, ":", sep="")
    print(f"  MSE: {mse}")
    print(f"  MAE: {mae}")
    print(f"  R2: {r2}")

Almacenamiento del modelo

# -----------------------------------------------------------------------------
# Ya no se requiere con MLflow
# -----------------------------------------------------------------------------
#
# def save_best_estimator(estimator):
#
#     import os
#     import pickle
#
#     if not os.path.exists("models"):
#         os.makedirs("models")
#     with open("models/estimator.pickle", "wb") as file:
#         pickle.dump(estimator, file)
#

Carga del modelo

# -----------------------------------------------------------------------------
# Ya no se requiere con MLflow
# -----------------------------------------------------------------------------
#
# def load_best_estimator():
#
#     import os
#     import pickle
#
#     if not os.path.exists("models"):
#         return None
#     with open("models/estimator.pickle", "rb") as file:
#         estimator = pickle.load(file)
#
#     return estimator
#

Entrenamiento

def train_estimator(alphas, l1_ratios, n_splits=5, verbose=1):

    import mlflow.sklearn
    from sklearn.linear_model import ElasticNet
    from sklearn.model_selection import GridSearchCV

    import mlflow

    x, y = load_data()
    x_train, x_test, y_train, y_test = make_train_test_split(x, y)

    # -------------------------------------------------------------------------
    # Búsqueda de parámetros con validación cruzada
    #
    estimator = GridSearchCV(
        estimator=ElasticNet(
            random_state=12345,
        ),
        param_grid={
            "alpha": alphas,
            "l1_ratio": l1_ratios,
        },
        cv=n_splits,
        refit=True,
        verbose=0,
        return_train_score=False,
    )
    # -------------------------------------------------------------------------

    estimator.fit(x_train, y_train)

    estimator = estimator.best_estimator_

    mse, mae, r2 = eval_metrics(y_test, y_pred=estimator.predict(x_test))
    if verbose > 0:
        report(estimator, mse, mae, r2)

    with mlflow.start_run():

        params = estimator.get_params()

        mlflow.log_param("alpha", params["alpha"])
        mlflow.log_param("l1_ratio", params["l1_ratio"])
        mlflow.log_metric("mse", mse)
        mlflow.log_metric("mae", mae)
        mlflow.log_metric("r2", r2)

        mlflow.sklearn.log_model(estimator, "model")

import numpy as np

train_estimator(
    alphas=np.linspace(0.0001, 0.5, 10),
    l1_ratios=np.linspace(0.0001, 0.5, 10),
    n_splits=5,
    verbose=1,
)

ElasticNet(alpha=0.0001, l1_ratio=0.16673333333333332, random_state=12345):
  MSE: 0.4555342217989927
  MAE: 0.5292810437627216
  R2: 0.34641239384627454

MLflow ui

Para visualizar la interfase use:

mlflow ui

Nota: En docker usar:

mlflow ui --host 0.0.0.0

con:

http://127.0.0.1:5001

Detalles de la corrida

Chequeo

def check_estimator():

    import mlflow

    x, y = load_data()
    x_train, x_test, y_train, y_test = make_train_test_split(x, y)

    # NOTA: este parámetro es copiado directamente de la interfase de MLflow
    estimator_path = "runs:/3aaec3b8a3724b3d8ca3406a98f47a1a/model"
    estimator = mlflow.pyfunc.load_model(estimator_path)
    mse, mae, r2 = eval_metrics(y_test, y_pred=estimator.predict(x_test))
    report(estimator, mse, mae, r2)


#
# Debe coincidir con el mejor modelo encontrado en la celdas anteriores
#
check_estimator()

%%bash
rm -rf outputs mlruns models