def _choose_classifier(job: Job):
method, config = get_method_config(job)
config.pop('classification_method', None)
logger.info("Using method {} with config {}".format(method, config))
if method == ClassificationMethods.KNN.value:
classifier = KNeighborsClassifier(**config)
elif method == ClassificationMethods.RANDOM_FOREST.value:
classifier = RandomForestClassifier(**config)
elif method == ClassificationMethods.DECISION_TREE.value:
classifier = DecisionTreeClassifier(**config)
elif method == ClassificationMethods.XGBOOST.value:
classifier = XGBClassifier(**config)
elif method == ClassificationMethods.MULTINOMIAL_NAIVE_BAYES.value:
classifier = MultinomialNB(**config)
elif method == ClassificationMethods.ADAPTIVE_TREE.value:
classifier = HAT(**config)
elif method == ClassificationMethods.HOEFFDING_TREE.value:
classifier = HoeffdingTree(**config)
elif method == ClassificationMethods.SGDCLASSIFIER.value:
classifier = SGDClassifier(**config)
elif method == ClassificationMethods.PERCEPTRON.value:
classifier = Perceptron(**config)
elif method == ClassificationMethods.NN.value:
config['encoding'] = job.encoding.value_encoding
config['is_binary_classifier'] = _check_is_binary_classifier(
job.labelling.type)
classifier = NNClassifier(**config)
else:
raise ValueError("Unexpected classification method {}".format(method))
return classifier
def _choose_classifier(job: Job):
if job.type == JobTypes.UPDATE.value:
classifier = _load_model(job.incremental_train)
# TODO: check if this instruction still makes sense
# are we updating a predictive_model with its own methods?
assert classifier[0].__class__.__name__ == job.method
else:
method, config = get_method_config(job)
config.pop('classification_method', None)
print("Using method {} with config {}".format(method, config))
if method == ClassificationMethods.KNN.value:
classifier = KNeighborsClassifier(**config)
elif method == ClassificationMethods.RANDOM_FOREST.value:
classifier = RandomForestClassifier(**config)
elif method == ClassificationMethods.DECISION_TREE.value:
classifier = DecisionTreeClassifier(**config)
elif method == ClassificationMethods.XGBOOST.value:
classifier = XGBClassifier(**config)
elif method == ClassificationMethods.MULTINOMIAL_NAIVE_BAYES.value:
classifier = MultinomialNB(**config)
elif method == ClassificationMethods.ADAPTIVE_TREE.value:
classifier = HAT(**config)
elif method == ClassificationMethods.HOEFFDING_TREE.value:
classifier = HoeffdingTree(**config)
elif method == ClassificationMethods.SGDCLASSIFIER.value:
classifier = SGDClassifier(**config)
elif method == ClassificationMethods.PERCEPTRON.value:
classifier = Perceptron(**config)
elif method == ClassificationMethods.NN.value:
config['encoding'] = job.encoding.value_encoding
config['is_binary_classifier'] = _check_is_binary_classifier(job.labelling.type)
classifier = NNClassifier(**config)
else:
raise ValueError("Unexpected classification method {}".format(method))
return classifier
def _choose_time_series_predictor(job: Job) -> TimeSeriesPredictorMixin:
method, config = get_method_config(job)
config.pop('time_series_prediction_method', None)
print("Using method {} with config {}".format(method, config))
if method == TimeSeriesPredictionMethods.RNN.value:
config['encoding'] = job.encoding.value_encoding
time_series_predictor = RNNTimeSeriesPredictor(**config)
else:
raise ValueError("Unexpected time series prediction method {}".format(method))
return time_series_predictor
def test_get_method_config(self):
job = create_test_job(
predictive_model=create_test_predictive_model(
predictive_model=PredictiveModels.CLASSIFICATION.value,
prediction_method=ClassificationMethods.RANDOM_FOREST.value
)
)
method, config = get_method_config(job)
self.assertEqual(ClassificationMethods.RANDOM_FOREST.value, method)
self.assertEqual({
'max_depth': None,
'max_features': 'auto',
'n_estimators': 10,
}, config)
def _choose_regressor(job: Job) -> RegressorMixin:
method, config = get_method_config(job)
config.pop('regression_method', None)
print("Using method {} with config {}".format(method, config))
if method == RegressionMethods.LINEAR.value:
regressor = LinearRegression(**config)
elif method == RegressionMethods.RANDOM_FOREST.value:
regressor = RandomForestRegressor(**config)
elif method == RegressionMethods.LASSO.value:
regressor = Lasso(**config)
elif method == RegressionMethods.XGBOOST.value:
regressor = XGBRegressor(**config)
elif method == RegressionMethods.NN.value:
config['encoding'] = job.encoding.value_encoding
regressor = NNRegressor(**config)
else:
raise ValueError("Unexpected regression method {}".format(method))
return regressor