Skip to content

Latest commit

 

History

History
88 lines (62 loc) · 5.04 KB

README.md

File metadata and controls

88 lines (62 loc) · 5.04 KB

StellarMapper

StellarMapper is a deep learning library created to provide easy and customizable access to SOTA deep learning architectures that might not always have provided implementations by the authors. At this moment, StellarMapper is focused on DL models that are introduced in various articles in astronomy and astrophysics journals, which often aren't packaged with a detailed explanation of their novel DL methodologies.

The library will also provide

  • utility tools for visualisation and interpretation of the various kinds of data used in astronomy (where solutions are not already available) as well as tools for hyperparameter tuning and model selection.
  • in-built model performance monitoring as well as several evaluation metrics.

Usage

Clone the repository (will add Pypi support when the library becomes extensive enough):

$ git clone https://github.com/bananya-ml/StellarMapper
$ cd ./stellarmapper

Create a virtual environment:

$ python -m venv .venv --upgrade-deps

where ".venv" is the name of your virtual environment. Activate the environment and install the required dependencies:

$ .\.venv\Scripts\activate
$ pip install -r ./requirements.txt

Navigate to the directory with the training examples:

$ cd ./src/training

and follow the notebook on how to set-up the model according to your data and needs.

  • NOTE Currently, I have not tested if the dependencies will hold up on Linux and MacOS machines. Quite likely some changes will be required to install the dependencies on anything other than Windows, so I am also including a rough list of the dependencies:
    • astropy==6.0.0
    • numpy==1.26.2
    • matplotlib==3.8.2
    • pandas==2.1.3
    • torch==2.1.2+cu121

Data

The datasets directory contains organized subdirectories named after the origin of the data. Each subdirectory contains a separate README detailing how the data was obtained and its usage.

Implemented Models

Model Title
StarNet An application of deep learning in the analysis of stellar spectra1
STARCNET StarcNet: Machine Learning for Star Cluster Identification2
OTRAIN O’TRAIN: A robust and flexible ‘real or bogus’ classifier for the study of the optical transient sky3
AlexNet One weird trick for parallelizing convolutional neural networks4
VGG Very Deep Convolutional Networks for Large-Scale Image Recognition5
  • NOTE The library contains some redundant architectures like 'AlexNet' and 'VGG' that were added for the sake of having accessibility within the same library. These models are quite popular and might have their official implementations in other libraries (e.g. torchvision).

TODO

❌ Add support for time series data
❌ Add (extensive) evaluation metrics
✅ Feature to save and load trained models

License

License: MIT

References