Local Normalization

Introduction

Local normalization is a module that normalizes the features of the local data (so that the p-norm ( p-norm ) of each feature after normalization takes a value of 1).

Specifically, for the feature matrix \(X\), perform Local Normalization transformation to obtain a new feature matrix \(\tilde{X}\). If features are normalized（axis=0）, then \(||\tilde{X}_{.j}||_p = 1\text{, }\forall j \in {1,\dots, m}\). If samples are normalized, then \(||\tilde{X}_{i.}||_p = 1\text{, }\forall i \in {1,\dots, n}\).

Parameter List

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

model_info:

  • name: str Model name, should be local_normalization.

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 Validation dataset type, currently supported is csv.

    • path: str The folder path of validation dataset.

    • name: str The file name of validation dataset.

    • has_id: bool Whether the dataset has id column.

    • has_label: bool Whether the dataset has label column.

output:

  • path: str Output folder path.

  • model:

    • name: str File name of output model.

  • trainset:

    • name: str File name of output train dataset.

  • valset:

    • name: str File name of output validation dataset.

train_info:
  • train_params:

    • norm: str The norm to use ("l1"/"l2"/"max").

    • axis: int Axis along which the normalization is applied. 1 for row normalization, 0 for column normalization.

    • feature_norm: map Fine-grained configuration for column normalization(axis = 0). The format is defined as: {column_1: {"norm": norm_1}, ..., column_N: {"norm": norm_N}}.