Vertical Feature Selection

Introduction

Feature selection relies on the output of algorithm VerticalBinningWoeIV and VerticalPearson.

The operator consists of three stages:

1. Vertical Binning Woe IV

2. Vertical Pearson

3. Performing feature selection based on the results of the above two steps.

Parameters List

The parameters given below is only for feature selection.

identity: str The role of each participant in federated learning, should be label_trainer or trainer.

model_info:

• name: str Model name, should be vertical_feature_selection.

input:

• trainset:

  • type: str Train dataset type, currently supported is csv.

  • path: str The folder path of train dataset.

  • name: str The file name of train dataset.

  • has_id: bool Whether the dataset has id column.

  • has_label: bool Whether the dataset has label column.

• valset:

  • type: str Train dataset type, currently supported is csv.

  • path: str The folder path of train dataset.

  • name: str The file name of train dataset.

  • has_id: bool Whether the dataset has id column.

  • has_label: bool Whether the dataset has label column.

• iv_result:

  • path: str Folder path of the iv_result from VerticalBinningWoeIV.

  • name: str Model file name.

• corr_result:

  • path: str Folder path of the result from VerticalPearson.

  • name: str Model file name.

output:

• path: str Output folder path.

• model:

  • name: str File name of output model.

• trainset:

  • name: str File name of train dataset after feature selection.

• valset:

  • name: str File name of validation dataset after feature selection.

train_info:

• train_params:

  • filter:

    • common:

      • metrics: str Metric type, currently supported is iv.

      • filter_method: str Feature filtering method, currently supported is threshold.

      • threshold: float Threshold for filtering if the filter_method is threshold.

    • correlation:

      • sort_metric: str Sorting (descending) metric before the correlated filtering, currently supported is iv.

      • correlation_threshold: float Threshold for correlated filtering.