Python-package Introduction

This document gives a basic walk-through of MVBLS Python-package.

List of other helpful links

• Python API

• Parameters Tuning

• Python Examples

Install

The preferred way to install MVBLS is via pip from Pypi:

pip install MVBLS

To verify your installation, try to import MVBLS in Python:

import MVBLS

Data Interface

The MVBLS Python module can load data from:

• NumPy 2D array(s) for single-view

  import numpy as np
  data = np.random.rand(500, 10)
  label = np.random.randint(2, size=500)
  

  Note: if input only one view data to MVBLS, it works like the original BLS [2].

• Dictionary for multi-view

  import numpy as np
  data={}
  data['view1'] = np.random.rand(500, 10)
  data['view2'] = np.random.rand(500, 10)
  data['view3'] = np.random.rand(500, 10)
  label = np.random.randint(2, size=500)
  

Setting Parameters

MVBLS can use a dictionary to set parameters. For instance:

• For single-view:

  param = {'view_list': None, 'reg_lambda': 0.01, 'n_nodes_Z':50, 'n_nodes_H': 6000, 'n_groups_Z': 10, 'reg_alpha': 0.01,
                   'random_state': 0}
  

• For multi-view:

  param = {'view_list': data.keys(), 'reg_lambda': 0.01, 'n_nodes_Z':50, 'n_nodes_H': 6000, 'n_groups_Z': 10, 'reg_alpha': 0.01,
                   'random_state': 0}
  

Training

Training a model requires a parameter dictionary and data set:

estimator = MVBLS.MVBLSClassifier(**param).fit(data, label)

After training, the model can be saved:

estimator.save_model('model.joblib')

A saved model can be loaded:

import joblib
estimator = joblib.load('model.joblib')

Predicting

A model that has been trained or loaded can perform predictions on datasets:

• For single-view:

  # 7 entities, each contains 10 features
  data = np.random.rand(7, 10)
  ypred = estimator.predict(data)
  

• For multi-view:

  # 7 entities, each contains 10 features
  data={}
  data['view1'] = np.random.rand(7, 10)
  data['view2'] = np.random.rand(7, 10)
  data['view3'] = np.random.rand(7, 10)
  ypred = estimator.predict(data)