NEEDLE2.0

NEEDLE2.0 is a classifier for building datasets and training transient-object classifiers with the needle_train pipeline.

Quick Start

python -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -e .

1. Configure paths in config.py:

   • DEFAULT_DATA_PATH
   • MAG_OUTPUT_PATH
   • HOST_DATA_PATH
   • OBJ_INFO_PATH

2. Build train/validation datasets:

python needle_train/get_train_valid_sets.py

3. Train and evaluate model:

python needle_train/run_model.py

Project Structure

• needle_train/: training, preprocessing, and dataset build scripts
• image/, light_curve/: image and photometry-related processing modules
• config.py: central configuration for paths and training options
• utils.py: shared helper functions

Inputs and Outputs

• Inputs: external data roots configured in config.py, with optional reference tables under info/.
• Outputs:
  • needle_inputs/ (NEEDLE_SET_PATH)
  • light_curve/photo_processing_output_new/ (PHOTO_OUTPUT_PATH)
  • image/image_preprocessing_output/ (IMG_OUTPUT_PATH)
  • image/image_unmasked_output/ (UNMASKED_IMG_OUTPUT_PATH)
• Large generated outputs are ignored by .gitignore to keep commits small.

License

This project is licensed under the MIT License. See LICENSE for details.