https://github.com/jafagervik/tinydas
Tinygrad meets DAS data
https://github.com/jafagervik/tinydas
ai autoencoder das ml tinygrad
Last synced: about 1 year ago
JSON representation
Tinygrad meets DAS data
- Host: GitHub
- URL: https://github.com/jafagervik/tinydas
- Owner: Jafagervik
- License: mit
- Created: 2024-07-23T09:13:28.000Z (almost 2 years ago)
- Default Branch: main
- Last Pushed: 2025-03-20T09:44:44.000Z (over 1 year ago)
- Last Synced: 2025-03-20T10:35:09.689Z (over 1 year ago)
- Topics: ai, autoencoder, das, ml, tinygrad
- Language: Python
- Homepage:
- Size: 190 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# TinyDAS - Tinygrad meets the PubDAS dataset
## How to get data
[See this pdf](https://dev.iris.edu/hq/files/initiatives/das_rcn/DAS-RCN-2022_12_02-Spica.pdf)
Or sign up to GLOBUS and go [here](https://app.globus.org/file-manager?origin_id=706e304c-5def-11ec-9b5c-f9dfb1abb183&origin_path=%2FFORESEE%2F&two_pane=false)
## Modules
The following is an explanation of the project structure
### Dataset
Loads the HDF5 data in from the `data` folder and exports it to a pytorch esc dataset
### Dataloader
Uses parallel workers to load single datafiles in parallel
### Models
See examples in the tinydas/models folder
All autoencoders are based on the BaseAE class
### Finding anomalies
Will upload jupyter notebooks soon
### Hyperparameters
They are stored in yaml files under the `configs` directory.
Name of the config is the name of the model in lowercase
## How to run
`python main.py -t train -m ae`
or alternatively
`python main.py -t detect -m ae`
# NOTES:
* Utils for loss scaling and clipping exist in this repo, but is kinda wonky for training certain models. However, F16 inference is easy:
1. Select model
2. Load model
3. model.half()