Boosting method implementation (LightGBM)

[RomanKharkovskoy]

План

• [ ] Реализовать convert_to_dataset()
• [ ] Обновить default_operation_params.json
• [ ] Обновить search_space.py

Как работает

Реализован интерфейс fit/predict в родительском классе FedotLightGBMtImplementation

Код

class FedotLightGBMImplementation(ModelImplementation):
    __operation_params = ['n_jobs', 'use_eval_set']

    def __init__(self, params: Optional[OperationParameters] = None):
        super().__init__(params)

        self.model_params = {k: v for k, v in self.params.to_dict().items() if k not in self.__operation_params}
        self.model = None

    def fit(self, input_data: InputData):
        input_data = input_data.get_not_encoded_data()

        if self.params.get('use_eval_set'):
            train_input, eval_input = train_test_data_setup(input_data)

            train_input = self.convert_to_dataframe(train_input)
            eval_input = self.convert_to_dataframe(eval_input)

            train_x, train_y = train_input.drop(columns=['target']), train_input['target']
            eval_x, eval_y = eval_input.drop(columns=['target']), eval_input['target']

            if self.classes_ is None:
                eval_metric = 'rmse'
            elif len(self.classes_) < 3:
                eval_metric = 'auc'
            else:
                eval_metric = 'multi_logloss'

            self.model.fit(X=train_x, y=train_y,
                           eval_set=[(eval_x, eval_y)], eval_metric=eval_metric)

        else:

            train_data = self.convert_to_dataframe(input_data)
            train_x, train_y = train_data.drop(columns=['target']), train_data['target']
            self.model.fit(X=train_x, y=train_y)

        return self.model

    def predict(self, input_data: InputData):
        input_data = self.convert_to_dataframe(input_data.get_not_encoded_data())
        train_x = input_data.drop(columns=['target'])
        prediction = self.model.predict(train_x)

        return prediction

Интерфейс fit/predict не поддерживает работу с внутренним типом данных lightgbm.Dataset, поэтому необходимо было найти обходной путь. В данном случае был использован тип данных pandas.DataFrame.

Внутри интерфейса идёт преобразование InputData в pandas.DataFrame (categorical_idx становятся category, а numerical_idx становятся float)

Код

@staticmethod
def convert_to_dataframe(data: Optional[InputData]):
    dataframe = pd.DataFrame(data=data.features, columns=data.features_names)
    dataframe['target'] = data.target

    if data.categorical_idx is not None:
        for col in dataframe.columns[data.categorical_idx]:
            dataframe[col] = dataframe[col].astype('category')

    if data.numerical_idx is not None:
        for col in dataframe.columns[data.numerical_idx]:
            dataframe[col] = dataframe[col].astype('float')

    return dataframe

[docu-mentor[bot]]

👋 Hi, I'm @docu-mentor, an LLM-powered GitHub app powered by Anyscale Endpoints that gives you actionable feedback on your writing.

Simply create a new comment in this PR that says:

@docu-mentor run

and I will start my analysis. I only look at what you changed in this PR. If you only want me to look at specific files or folders, you can specify them like this:

@docu-mentor run doc/ README.md

In this example, I'll have a look at all files contained in the "doc/" folder and the file "README.md". All good? Let's get started!

[github-actions[bot]]

All PEP8 errors has been fixed, thanks :heart:

Comment last updated at

[valer1435]

@open-code-helper run

[valer1435]

@open-code-helper run

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 77.88462% with 23 lines in your changes missing coverage. Please review.

Project coverage is 80.16%. Comparing base (80eba8e) to head (fd3786d). Report is 1 commits behind head on master.

Files	Patch %	Lines
...mplementations/models/boostings_implementations.py	77.88%	23 Missing :warning:

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1264      +/-   ##
==========================================
+ Coverage   80.10%   80.16%   +0.05%     
==========================================
  Files         146      146              
  Lines       10190    10284      +94     
==========================================
+ Hits         8163     8244      +81     
- Misses       2027     2040      +13     

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.