hivae 0.1

HIVAE (https://arxiv.org/pdf/1807.03653.pdf - by Nazabal, Olmos, Ghahramani, Valera) - extenstion of their implementations as Python library

hivae

This repository contains the Modular reimplemenation of the Heterogeneous Incomplete Variational Autoencoder model (HI-VAE)written by Alfredo Nazabal ([e-mail]). It was written in Python, using Tensorflow.

The details of this model are included in this paper. Please cite it if you use this code/library for your own research.

Database description

There are three different example datasets found in the library (Wine, Adult and Diabetes). Majority of the datasets( Wine and Adult) have each their own folder, containing:

• data.csv: the dataset
• data_types.csv(NOT REQUIRED, LOOK AT THE EXAMPLE(working_example): a csv containing the types of that particular dataset. Every line is a different attribute containing three paramenters:
  • type: real, pos (positive), cat (categorical), ord (ordinal), count
  • dim: dimension of the variable
  • nclass: number of categories (for cat and ord)
• Missingxx_y.csv: a csv containing the positions of the different missing values in the data. Each "y" mask was generated randomly, containing a "xx" % of missing values.

You can add your own datasets as long as they follow this structure.

Files description

• HIVAE.py: The main script of the library, it needs to imported to work with the library and is connected to all the other scripts.
• loglik_ models_ missing_normalize.py: In this file, the different likelihood models for the different types of variables considered (real, positive, count, categorical and ordinal) are included.
• model_ HIVAE_inputDropout.py: Contains the HI-VAE with input dropout encoder model.
• model_ HIVAE_factorized.py: Contains the HI-VAE with factorized encoder model

Contact

• For questions regarding algorithm --> Alfredo Nazabal: [e-mail]
• For bugs or suggestion regarding code --> Fathy Shalaby: [e-mail]

Comments on general_example.py

main_directory: where is the project folder

dataset_name: the name of the database (if you want)

types_list_d: a dictionary where the key is the dataset name, which contains a list with tuples that indicates the column names, types, the number of dimensions and classes

types:

• count: real values

• cat: categorical 0 or 1

• pos: positive real values

• ordinal: ordinal number

number of dimensions:

• number of possibilities in the categorical variables or 1 in numerical

number of classes:

• number of options (same of number of dimensions for categorical variables)

dataset_path: this is the folder of the csv files

results_path: the output folder for results

network_path: where the models are going to be stored

types_list: the specific type for the dataset you are going to use data_file: the full dataset train_file/ test_file: if the dataset was already splitted

train_data/test_data: pandas dataframes

dim_y: the depth of the network

dim_s/dim_z: dimensions of the embedding