Alien Megastructure Transit Reconstruction

This repository contains the research code for generating synthetic transit light curves from arbitrary occulting shapes and training a neural network to reconstruct a 2D shadow image from a 1D light curve.

The project is organized as a pipeline:

1. Generate 2D shapes that represent transiting objects.
2. Simulate light curves for those shapes with EightBitTransit.
3. Convert raw images and CSV light curves into .npy datasets.
4. Train a TensorFlow model that maps a processed light curve to a reconstructed shape.
5. Run inference on synthetic and real-world-inspired examples such as BATMAN, Kepler, and exocomet light curves.

Datasets and trained model artifacts referenced by the original experiments are stored externally:

• Google Drive: https://drive.google.com/drive/folders/15Cz5QkND9pC6HosmHSYo_b3DK3SVAEVp?usp=sharing

Start Here

• Documentation hub: docs/README.md
• Data generation pipeline: docs/data-generation.md
• Model architecture: docs/model.md
• Training workflow: docs/training.md
• Inference workflow: docs/inference.md

Repository Layout

• lightcurve_simulation/: shape generation, raw light-curve simulation, and conversion utilities.
• ml_training/: training and inference scripts, Colab notebooks, and archived experiments.
• data/: dataset metadata and experiment notes used by the project.
• lc_fitting/: notebook-based fitting experiments.

What The Current Code Assumes

This is a research codebase rather than a packaged library. Many scripts:

• use hard-coded absolute paths such as /scratch/...,
• prompt for confirmation with input("Do you want to run the code?"),
• expect datasets to already exist on local disk,
• were written as one-off experiment runners instead of reusable CLIs.

The new docs/ folder explains how those pieces fit together and points to the main scripts to use first.

Main Entry Points

• Shape generation: lightcurve_simulation/data_raw/create_shapes.py
• Light-curve simulation for one .npy shape file: lightcurve_simulation/data_raw/pll_create_8bit_lc.py
• Light-curve simulation for batched training shapes: lightcurve_simulation/data_raw/frontend_rand_param_batch_lc_gen.py
• Convert raw light curves to .npy: lightcurve_simulation/data_npy/cnvt_lcFlux_to_npy.py
• Training script: ml_training/ml_train_lightcurve_unf_dist_LDC_size.py
• Python inference script: ml_training/ml_inference_lc2shape.py
• Final notebook workflow: ml_training/Google_Colab_Codes/megastructure_ml_model_training.ipynb

Recommended Reading Order For New Contributors

1. Read docs/README.md for the big picture.
2. Read docs/data-generation.md to understand the dataset.
3. Read docs/model.md and docs/training.md.
4. Use docs/inference.md to interpret the inference notebooks and script outputs.