How pablovd/cosmographnet works
CosmoGraphNet turns CAMELS galaxy catalogues into graphs, then trains or evaluates a graph neural network for either cosmological-parameter inference or galaxy power-spectrum prediction. The main workflow is configuration → graph-dataset construction and splitting → GNN → training/testing → saved predictions and plots. Separate capabilities include hyperparameter optimization, cross-suite evaluation, power-spectrum extrapolation tests, and graph visualization. The README documents the latter test workflows, but their implementation details were not sampled; those connections are therefore left unverified.
Main entry points: main.py Run orchestrator · training.py Train and test · visualize_graphs.py Graph visualization · load_data.py Graph construction
Components 18
Training and inference
- Run configuration
hyperparameters.py - Run orchestrator
main.py - Train and test
Source/training.py - Saved models
Models - Shared runtime settings
Source/constants.py
Graph model
- Graph network
Source/metalayer.py - Message passing
Source/metalayer.py
Evaluation and analysis
- Prediction outputs
- Result plotting
Source/plotting.py - Hyperparameter search
hyperparams_optimization.py - Cross-suite test
crosstest.py - Spectrum extrapolation
ps_test.py - Graph visualization
visualize_graphs.py
Simulation data
- Graph construction
Source/load_data.py - Dataset splitting
Source/load_data.py - CAMELS catalogues
- Power spectra
PS_files
Other components
- Researcher
Connections 22
- Researcher → Run orchestrator: runs
README.md - Run configuration → Run orchestrator: configures
main.py - Run orchestrator → Graph construction: creates dataset
main.py - Graph construction → CAMELS catalogues: reads catalogues
Source/load_data.py - Graph construction → Power spectra: reads targets
Source/load_data.py - Run orchestrator → Dataset splitting: splits dataset
main.py - Run orchestrator → Graph network: initializes
main.py - Graph network → Message passing: runs layers
Source/metalayer.py - Run orchestrator → Train and test: trains and tests
main.py - Train and test → Graph network: forwards batches
Source/training.py - Train and test → Saved models: saves weights
Source/training.py - Run orchestrator → Saved models: loads weights
main.py - Train and test → Prediction outputs: saves predictions
Source/training.py - Run orchestrator → Result plotting: plots results
main.py - Result plotting → Prediction outputs: reads predictions
Source/plotting.py - Hyperparameter search → Run orchestrator: runs trials
hyperparams_optimization.py - Cross-suite test → Run orchestrator: evaluates suites
crosstest.py - Researcher → Spectrum extrapolation: runs evaluation
README.md - Researcher → Graph visualization: visualizes graphs
README.md - Graph visualization → CAMELS catalogues: reads catalogues
visualize_graphs.py - Graph visualization → Result plotting: uses plotting helpers
visualize_graphs.py - Run orchestrator → Shared runtime settings: uses runtime settings