One-Hot Encoder FeatureCloud App

Allow encoding categorical columns with M categories into M-1 columns according to the one-hot scheme. The order of non-categorical columns is preserved, encoded columns are inserted inplace of the original column.

Modes

The app can run in two modes; auto and predefined.

• In the auto mode categorical columns and the possible discrete values are detected automatically. This may leak private data and open attack vectors for non-trusted nodes, but requires no additional setup.

• In the predefined mode the coordinator must define categorical columns and their values beforehand. To use this, add the categorical_variables directive in your config.yml. Under this directive define key-value-pairs for each categorical column in your data. These should consist of a string denoting the column name and a list enumerating all possible values. If the data contains a not listed value, the row will be dropped.

The nodes must agree on the selected mode. If the modes are inconsistent, execution will be stopped prematurely before exchanging further data.

Example configs for the predefined mode

At coordinator:

fc_one_hot_encoding:
  files:
    input_filename: "client1.csv"
    output_filename: "client1_ohe.csv"
    sep: ","
  mode: "predefined"
  categorical_variables:
    age: [ '20-29', '30-39', '40-49', '50-59', '60-69', '70-79' ]
    menopause: [ 'ge40', 'lt40', 'premeno' ]
    tumor-size: ['0-4', '10-14', '15-19', '20-24', '25-29', '30-34', '35-39', '40-44', '45-49', '5-9', '50-54' ]
    inv-nodes: ['0-2', '12-14', '15-17', '24-26', '3-5', '6-8', '9-11']
    node-caps: ['no', 'yes']
    deg-malig: [1, 2, 3]
    breast: ['left', 'right']
    breast-quad: ['central', 'left_low', 'left_up', 'right_low', 'right_up']
    irradiat: ['no', 'yes']

At non-coordinator nodes:

fc_one_hot_encoding:
  files:
    input_filename: "client2.csv"
    output_filename: "client2_ohe.csv"
    sep: ","
  mode: "predefined"