Skip to content

Initial FAITH commit#29

Open
nathanchenseanwalter wants to merge 120 commits into
mainfrom
foundation25
Open

Initial FAITH commit#29
nathanchenseanwalter wants to merge 120 commits into
mainfrom
foundation25

Conversation

@nathanchenseanwalter
Copy link
Copy Markdown
Collaborator

No description provided.

renierts and others added 30 commits April 18, 2024 08:59
- Added .gitignore
- Added a directory for documentation.
- Added a directory for unit tests.
With a script of fetching data, transferring data, and object (work-in-progress) that read and unify the dataset.
Work Time series full pipeline take suffix, output aligned dict with window.
Minor improvement on the finding the closest index.
- Added spectrogram utilities.
- Further cosmetic changes in Max's code
# Conflicts:
#	examples/Data_fetching/fetch_GAdata.py
#	examples/Data_fetching/fetch_toksearch.py
Added functionalities to resample a time-series and an empty module for time-series interpolation.
Co-authored-by: Alvin Garcia <alvin-garcia@users.noreply.github.com>
…rom "fusion_ai_hub" to "fusionaihub". Removed several unused modules and files related to data loading, processing, and visualization. Added new dependencies for enhanced functionality, including ipykernel, ipywidgets, scikit-learn, torch, tables, and pyyaml.
…r notebooks to reset execution counts, streamline data loading, and enhance visualization. Refactor dataset preparation scripts to improve logging, configuration handling, and remove deprecated modules.
…nvironment and installing necessary packages.
…uctions. Remove deprecated logging configuration file and update dataset preparation README for improved clarity and modular structure.
…ormats from .pkl to .joblib and .csv for dataset indexing, enhancing clarity on the output structure.
nathanchenseanwalter and others added 19 commits July 19, 2025 10:08
Added instructions on how to use
- Increased job execution time in `prepare_data.sh` from 10 hours to 15 hours.
- Changed the data processing configuration in `accessing_data.ipynb` to use the `magnetics` dataset instead of `signals`.
- Enabled debug mode in `spectrogram.yaml` for improved troubleshooting.
- Updated STFT transformation parameters in `processing_v0.py` to utilize configuration settings.
- Enhanced type hints in dataset classes for better clarity and type safety in `base.py`, `file_based.py`, and other related files.
- Added missing newlines at the end of files in `__main__.py` and `align.py`.
- Improved formatting and consistency in `base.py`, `file_based.py`, and other dataset classes by adjusting indentation and line breaks for better readability.
- Added `black[jupyter]` to development dependencies in `pyproject.toml` for improved code formatting.
- Introduced new ignore rule for `black` in `pyproject.toml` to handle specific annotation warnings.
- Updated function signatures in various files to include type hints for better clarity and type safety.
- Improved formatting in several Python files for consistency and readability.
- Added notes in `README.md` regarding package reinstallation after changes.
- Removed obsolete Jupyter notebooks from the `hackathon` directory to clean up the project.
Big Beautiful Bill: 1300 linting error fixes + Ruff + Black reformatting + fixes to dataloader (can now read joblib dataset)
This commit also contains some updates to the pinned dependencies in the
`uv.lock` file.  It might be a good idea to re-visit the dependencies
once we have made more progress on the project.
Implement draft spectrogram-based model `specfmv0`
@review-notebook-app
Copy link
Copy Markdown

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.


Powered by ReviewNB

Copy link
Copy Markdown

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull Request Overview

This is an initial commit introducing the FAITH (Fusion Autoencoder for Interpretable Token-based Hierarchical representations) framework for training block-based autoencoders. The PR includes a complete implementation of modular autoencoder architectures with support for masked autoencoders (MAE), hyperparameter tuning via Ray Tune, PyTorch Lightning integration, and flexible file-based datasets.

Key changes:

  • Modular autoencoder architecture with configurable encoder/decoder blocks
  • Masked Autoencoder (MAE) implementation with various masking strategies
  • Ray Tune integration for hyperparameter optimization with multiple search algorithms
  • PyTorch Lightning training framework with automatic mixed precision and scheduling
  • Flexible dataset loading supporting joblib, HDF5, and NumPy file formats

Reviewed Changes

Copilot reviewed 101 out of 129 changed files in this pull request and generated 8 comments.

Show a summary per file
File Description
tests/train/test_train_blocks_base.py Example usage and testing of base autoencoder blocks
tests/train/test_autoencoder.py Comprehensive testing of BlockBasedAutoencoder functionality
test_config.yaml YAML configuration example for autoencoder models
src/faith/train/tuning/search_spaces.py Predefined hyperparameter search spaces for Ray Tune
src/faith/train/tuning/ray_tuner.py Ray Tune integration for hyperparameter optimization
src/faith/train/tuning/__init__.py Tuning module initialization with graceful Ray import handling
src/faith/train/training/lightning_trainer.py PyTorch Lightning wrapper for autoencoder training
src/faith/train/training/__init__.py Training module exports
src/faith/train/models/utils.py Utility functions for model analysis and memory estimation
src/faith/train/models/mae.py Masked Autoencoder implementation with flexible masking strategies
src/faith/train/models/configs.py Configuration management with preset and custom model configurations
src/faith/train/models/autoencoder.py Core BlockBasedAutoencoder implementation
src/faith/train/models/__init__.py Model module exports and public API
src/faith/train/data/loaders/factory.py DataLoader factory with worker initialization for lazy datasets
src/faith/train/data/datasets/file_based.py File-based dataset implementations supporting multiple formats

Comment thread src/faith/train/tuning/ray_tuner.py Outdated
Comment thread src/faith/train/tuning/ray_tuner.py Outdated
Comment thread src/faith/train/models/mae.py Outdated
Comment thread src/faith/train/data/datasets/file_based.py Outdated
Comment thread src/faith/train/data/datasets/file_based.py Outdated
Comment thread src/faith/train/models/autoencoder.py Outdated
Comment thread tests/train/test_autoencoder.py Outdated
Comment thread src/faith/train/models/__init__.py
nathanchenseanwalter and others added 7 commits August 1, 2025 15:32
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants