Model selection Module (API Reference)

Main Cross-validation class for adding more estimators and get score using CV.

This will contain some model selection logic should be used here like Grid search logic, as here what I could do is create the search space. So maybe Grid search to find whole models.

Classifier real training happens here.

@author: Guangqiang.lu

class automl.model_selection.GridSearchModel(backend, n_best_model=None, use_neural_network=True, task_type='classification')

Bases: object

Here could just to implement that could add a list of estimators and their parameters list. I want to make this class to do real training part.

self.estimator_list is like: [GridSearchCV(lr, params), …] self.score_dict is like: {‘LogisticRegressin’: (lr, 0.9877)}

add_estimator(estimator, estimator_params=None)

As I also want to keep current logic with a list of estimators in parallel, so here should keep whole models. As I will my own estimator, so don’t need always need estimator_params, but just add in case we just want to add other sklearn estimator :param estimator: estimator object :param estimator_params: native sklearn object params to search. :return:

add_estimators(estimator_param_pairs)

This is try to parallel training for whole training with different params. :param estimator_param_pairs: a list of estimator and params

just like this:[(lr, {“C”:[1, 2]})]

Returns

fit(x, y, n_jobs=None)

Fit and Cross-validation for each estimator with scoring supported!

After CV and re-fit with full data, also the training_score and testing_score is based on CV result. Support with processing information by tqdm.

By defualt:

classification scoring is accuracy regression scoring is mean_squared_error

Parameters

• x – training data

• y – training label

• n_jobs – how much cores to use

Returns

load_best_model_list(model_extension='pkl')

Load previous saved best model into a list of trained instance. :return:

load_bestest_model()

load best trained model from disk :return:

predict(x)

Get prediction based on best fitted model :param x: :return:

predict_proba(x)

Get probability of based on best estimator :param x: :return:

print_estimators()

To list whole grid models instances, so that we could check. :return:

save_best_model_list()

save whole best fitted model based on each algorithm own parameters, so that we could save each best models. Then we could do ensemble logic.

Here I think I could just save the each best parameters trained model into disk, also the file name should be like LogisticRegression_9813.pkl: with classname_score.pkl.

Noted: This func should only be called after trained :param n_best_model: How many best model to save :return:

save_bestest_model()

dump best trained model into disk, one more thing here: We shouldn’t save the model into disk with fixed name, but should with score. :return:

save_trained_estimator(estimator, estimator_name)

To save the trained model into disk with file name :param estimator: :param estimator_name: :return:

score(x, y)

To use best fitted model to evaluate test data :param x: :param y: :return: