hvae-backbone 0.1.100

Universal and customizable implementation of the Hierarchical Variational Autoencoder architecture.

hVAE - backbone

This repository contains:

• customizable backbone implementation of hVAE
• reuseable hVAE components such as blocks, layers, losses, etc.
• training, evaluation and analyzation scripts
• checkpoint handling (using Weights & Biases)

Installation

pip install hvae_backbone

Usage

This repository is intended to be used as a backend package.
Please refer to hvae template repository for usage instructions.

Project Structure

├── hvae_backbone
│   ├── elements
│   │   ├── __init__.py             
│   │   ├── data_preproc.py         # Modules for data preprocessing
│   │   ├── dataset.py              # Base dataset class
│   │   ├── distributions.py        # Distributions, distributions generation
│   │   ├── layers.py               # Layers for building models
│   │   ├── losses.py               # Loss functions
│   │   ├── nets.py                 # Network architectures
│   │   ├── optimizers.py           # Optimizers
│   │   ├── schedules.py            # Schedules e.g. LR, KL weight
│   ├── __init__.py                 # package level scripts
│   ├── analysis.py                 # Analysis tools for trained models
│   ├── block.py                    # Blocks for building hierarchical models
│   ├── checkpoint.py               # Checkpoint handling (save, load)
│   ├── functional.py               # Functional scripts (training, loss, etc.)
│   ├── hvae.py                     # General hVAE class
│   ├── sequence.py                 # General sequential hVAE class
│   ├── utils.py                    # Utility functions

TODO:

• callbacks
• preprocessing
• sample vs rsmaple blokkoknál (SimpleGenBlock)
• weight initialization