Data

This document provides explanation and source information on data used for running examples. Many of the data sets have been scaled or transformed from their original version.

Data Set Naming Rule

All data sets are named according to this guideline: name: “{content}_{mode}_{size}_{role}_{role_index}”

• content: brief description of data content

• mode: how original data is divided, either “homo””or hetero”; some data sets do not have this information

• size: includes keyword “mini” if the data set is truncated from another larger set

• role: role name, either “host” or “guest”

• role_index: if a data set is further divided and shared among multiple hosts in some example, indices are used to distinguish different parties, starts at 1

Data sets used for running examples are uploaded to FATE data storage at the time of deployment. Uploaded tables share the same namespace “experiment” and have table_name matching to original file names. Below lists example data sets and their information.

Horizontally Divided Data

For Homogeneous Federated Learning

breast_homo:

• 30 features

• label type: binary

• source

• data sets:

  1. “breast_homo_guest.csv”

     • name: “breast_homo_guest”

     • namespace: “experiment”

  2. “breast_homo_host.csv”

     • name: “breast_homo_host”

     • namespace: “experiment”

  3. “breast_homo_test.csv”

     • name: “breast_homo_test”

     • namespace: “experiment”

default_credit_homo:

• 23 features

• label type: binary

• source

• data sets:

  1. “default_credit_homo_guest.csv”

     • name: “default_credit_homo_guest”

     • namespace: “experiment”

  2. “default_credit_homo_host_1.csv”

     • name: “default_credit_homo_host_1”

     • namespace: “experiment”

  3. “default_credit_homo_host_2.csv”

     • name: “default_credit_homo_host_2”

     • namespace: “experiment”

  4. “default_credit_homo_test.csv”

     • name: “defeault_credit_homo_test”

     • namespace: “experiment”

epsilon_5k_homo:

• 100 features

• label type: binary

• mock data

• data sets:

  1. “epsilon_5k_homo_guest.csv”

     • name: “epsilon_5k_homo_guest”

     • namespace: “experiment”

  2. “epsilon_5k_homo_hostt.csv”

     • name: “epsilon_5k_homo_host”

     • namespace: “experiment”

  3. “epsilon_5k_homo_test.csv”

     • name: “epsilon_5k_homo_test”

     • namespace: “experiment”

give_credit_homo:

• 10 features

• label type: binary

• source

• data sets:

  1. “give_credit_homo_guest.csv”

     • name: “give_credit_homo_guest”

     • namespace: “experiment”

  2. “give_credit_homo_host.csv”

     • name: “give_credit_homo_host”

     • namespace: “experiment”

  3. “give_credit_homo_test.csv”

     • name: “give_credit_homo_test”

     • namespace: “experiment”

student_homo:

• 13 features

• label type: continuous

• source

• data sets:

  1. “student_homo_guest.csv”

     • name: “student_homo_guest”

     • namespace: “experiment”

  2. “student_homo_host.csv”

     • name: “student_homo_host”

     • namespace: “experiment”

  3. “student_homo_test.csv”

     • name: “student_homo_test”

     • namespace: “experiment”

vehicle_scale_homo:

• 18 features

• label type: multi-class

• source

• data sets:

  1. “vehicle_scale_homo_guest.csv”

     • name: “vehicle_scale_homo_guest”

     • namespace: “experiment”

  2. “vehicle_scale_homo_host.csv”

     • name: “vehicle_scale_homo_host”

     • namespace: “experiment”

  3. “vehicle_scale_homo_test.csv”

     • name: “vehicle_scale_homo_test”

     • namespace: “experiment”

Vertically Divided Data

For Heterogeneous Federated Learning

breast_hetero:

• 30 features

• label type: binary

• source

• data sets:

  1. “breast_hetero_guest.csv”

     • name: “breast_hetero_guest”

     • namespace: “experiment”

  2. “breast_hetero_host.csv”

     • name: “breast_hetero_host”

     • namespace: “experiment”

breast_hetero_mini:

• 7 features

• label type: binary

• source

• data sets:

  1. “breast_hetero_mini_guest.csv”

     • name: “breast_hetero_mini_guest”

     • namespace: “experiment”

  2. “breast_hetero_mini_host.csv”

     • name: “breast_hetero_mini_host”

     • namespace: “experiment”

default_credit_hetero:

• 23 features

• label type: binary

• source

• data sets:

  1. “default_credit_hetero_guest.csv”

     • name: “default_credit_hetero_guest”

     • namespace: “experiment”

  2. “default_credit_hetero_host.csv”

     • name: “default_credit_hetero_host”

     • namespace: “experiment”

dvisits_hetero:

• 12 features

• label type: continuous

• source

• data sets:

  1. “dvisits_hetero_guest.csv”

     • name: “dvisits_hetero_guest”

     • namespace: “experiment”

  2. “dvisits_hetero_host.csv”

     • name: “dvisits_hetero_host”

     • namespace: “experiment”

epsilon_5k_hetero:

• 100 features

• label type: binary

• mock data

• data sets:

  1. “epsilon_5k_hetero_guest.csv”

     • name: “epsilon_5k_hetero_guest”

     • namespace: “experiment”

  2. “epsilon_5k_hetero_host.csv”

     • name: “epsilon_5k_hetero_host”

     • namespace: “experiment”

give_credit_hetero:

• 10 features

• label type: binary

• source

• data sets:

  1. “give_credit_hetero_guest.csv”

     • name: “give_credit_hetero_guest”

     • namespace: “experiment”

  2. “give_credit_hetero_host.csv”

     • name: “give_credit_hetero_host”

     • namespace: “experiment”

ionosphere_scale_hetero

• 34 features

• label type: binary

• source

• data sets:

  1. “ionosphere_scale_hetero_guest.csv”

     • name: “ionosphere_scale_hetero_guest”

     • namespace: “experiment”

  2. “ionosphere_scale_hetero_host.csv”

     • name: “ionosphere_scale_hetero_host”

     • namespace: “experiment”

motor_hetero:

• 11 features

• label type: continuous

• source

• data sets:

  1. “motor_hetero_guest.csv”

     • name: “motor_hetero_guest”

     • namespace: “experiment”

  2. “motor_hetero_host.csv”

     • name: “motor_hetero_host”

     • namespace: “experiment”

  3. “motor_hetero_host_1.csv”

     • name: “motor_hetero_host_1”

     • namespace: “experiment”

  4. “motor_hetero_host_2.csv”

     • name: “motor_hetero_host_2”

     • namespace: “experiment”

motor_hetero_mini:

• 7 features

• label type: continuous

• source

• data sets:

  1. “motor_hetero_mini_guest.csv”

     • name: “motor_hetero_mini_guest”

     • namespace: “experiment”

  2. “motor_hetero_mini_host.csv”

     • name: “motor_hetero_mini_host”

     • namespace: “experiment”

student_hetero:

• 13 features

• label type: continuous

• source

• data sets:

  1. “student_hetero_guest.csv”

     • name: “student_hetero_guest”

     • namespace: “experiment”

  2. “student_hetero_host.csv”

     • name: “student_hetero_host”

     • namespace: “experiment”

vehicle_scale_hetero:

• 18 features

• label type: multi-class

• source

• data sets:

  1. “vehicle_scale_hetero_guest.csv”

     • name: “vehicle_scale_hetero_guest”

     • namespace: “experiment”

  2. “vehicle_scale_hetero_host.csv”

     • name: “vehicle_scale_hetero_host”

     • namespace: “experiment”

Federated Transfer Learning Data

For Federated Transfer Learning

nus_wide:

• 636/1000 features

• label type: binary

• source

• data sets:

  1. “nus_wide_train_guest.csv”

     • name: “nus_wide_train_guest”

     • namespace: “experiment”

  2. “nus_wide_train_host.csv”

     • name: “nus_wide_train_host”

     • namespace: “experiment”

  3. “nus_wide_validate_guest.csv”

     • name: “nus_wide_validate_guest”

     • namespace: “experiment”

  4. “nus_wide_validate_guest.csv”

     • name: “nus_wide_validate_guest”

     • namespace: “experiment”

Non-Divided Data

Generated Data for Data Operation Demo

tag_value:

• data sets:

  1. “tag_value_1000_140.csv”

     • name: “tag_value_1”, “tag_value_2”, “tag_value_3”

     • namespace: “experiment”