def transform()

in recommenders/datasets/pandas_df_utils.py [0:0]

• 38 lines of code
• 14 McCabe index (conditional complexity)

    def transform(self, df):
        """Tranform an input dataset with the same schema (column names and dtypes) to libffm format
        by using the fitted converter.

        Args:
            df (pandas.DataFrame): input Pandas dataframe.

        Return:
            pandas.DataFrame: Output libffm format dataframe.
        """
        if self.col_rating not in df.columns:
            raise ValueError(
                "Input dataset does not contain the label column {} in the fitting dataset".format(
                    self.col_rating
                )
            )

        if not all([x in df.columns for x in self.field_names]):
            raise ValueError(
                "Not all columns in the input dataset appear in the fitting dataset"
            )

        # Encode field-feature.
        idx = 1
        self.field_feature_dict = {}
        for field in self.field_names:
            for feature in df[field].values:
                # Check whether (field, feature) tuple exists in the dict or not.
                # If not, put them into the key-values of the dict and count the index.
                if (field, feature) not in self.field_feature_dict:
                    self.field_feature_dict[(field, feature)] = idx
                    if df[field].dtype == object:
                        idx += 1
            if df[field].dtype != object:
                idx += 1

        self.field_count = len(self.field_names)
        self.feature_count = idx - 1

        def _convert(field, feature, field_index, field_feature_index_dict):
            field_feature_index = field_feature_index_dict[(field, feature)]
            if isinstance(feature, str):
                feature = 1
            return "{}:{}:{}".format(field_index, field_feature_index, feature)

        for col_index, col in enumerate(self.field_names):
            df[col] = df[col].apply(
                lambda x: _convert(col, x, col_index + 1, self.field_feature_dict)
            )

        # Move rating column to the first.
        column_names = self.field_names[:]
        column_names.insert(0, self.col_rating)
        df = df[column_names]

        if self.filepath is not None:
            np.savetxt(self.filepath, df.values, delimiter=" ", fmt="%s")

        return df