Save CatBoost model Object from Optuna Objective

Hello,

I’m currently tuning a CatBoost model using the standard Optuna Objective function approach like so:

def objective(trial: optuna.Trial):
    
  ...
  ...
  ...

    
    # Initialsing CatBoost model
    model = cat.CatBoostClassifier(
        loss_function = "CrossEntropy",
        eval_metric = "AUC",
#         task_type = "GPU",
        **params)
    
    model.fit(training_set, early_stopping_rounds = 10)
    
    # Predicting
    prediction = model.predict(validation_set)
    prediction_label = np.rint(prediction)
    
    # Evaluating
    roc_auc = roc_auc_score(y_val, prediction_label)
    f1 = f1_score(y_val, prediction_label)
    recall = recall_score(y_val, prediction_label)
    
    print('ROC AUC Score:', roc_auc)
    print('F1 Score:', f1)
    print('Recall Score:', recall)
    
    return roc_auc, f1, recall

However, I’m trying to save the best CatBoost model that was trained and evaluated on my train and validation sets. I want to save the model as an object to then test on my test set. I appreciate Optuna is used for hyperparameter tuning but I wanted to know whether it was possible to just extract the best model directly from my objective function to avoid re-fitting a new model.

I’ve attempted to follow this example that achieves this using a LightGBM model. Is there a way to achieve this with a CatBoost instead? Thank you

[nzw0301]

I suppose that we can save catboost models by using save_model function rather than re-fitting.

[ziadzee]

Hi, @nzw0301 - thank you for the reply. It seems to work as intended when I convert the objective function to a single objective function, however not have much luck when I attempt with multiple objectives. In this case, I’ll just have to re-fit a new CatBoost model with the hyperparameters obtained from my multi-objective function.