Universal and customizable implementation of the Hierarchical Variational Autoencoder architecture.
Project description
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
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
hvae_backbone-0.1.48.tar.gz
(33.7 kB
view hashes)
Built Distribution
Close
Hashes for hvae_backbone-0.1.48-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 1bf63e7d12fa8b3ca152fd3c25c797cf49909e61785ff3023373c9c8ab100de7 |
|
MD5 | b2aca0df4fa22f699694bdc2c4e254d6 |
|
BLAKE2b-256 | 184353a944cb0ec65edf5c0c61454c119b239c6d7cfceb9a8015cf3184803ee7 |