How simonbing/crlsanitycheck works
This repository is an experimental benchmark for sanity-checking causal representation learning on Causal Chamber real data and synthetic or semi-synthetic data. The principal workflow is to select an experiment and dataset, prepare train/validation/test data, fit a method, and evaluate or save results. The graph preserves the documented Contrastive CRL, Multiview CRL, CITRIS, and supervised-fit entry points, plus the shared data/model layer and the Contrastive CRL implementation. Some experiment internals are unsampled, so only documented or indexed relationships are shown.
Main entry points: contrastive_crl_experiment.py Contrastive experiment · run_experiments.py Baseline runner · multiview_experiment.py Multiview experiment · data_generation.py Contrastive data
Components 18
Experiment entry points
- Contrastive experiment
crc/contrastive_crl_experiment.py - Multiview experiment
crc/multiview_experiment.py - CITRIS experiment
crc/citris_experiment.py - Supervised fit
crc/supervised_fit_experiment.py - Baseline runner
crc/baselines/contrastive_crl/src/run_experiments.py
Data and generation
- Chamber datasets
crc/shared/torch_datasets.py - Contrastive data
crc/baselines/contrastive_crl/src/data_generation.py - Data selection
crc/baselines/contrastive_crl/src/utils.py
CRL methods
- Contrastive models
crc/baselines/contrastive_crl/src/models.py - Nonlinearities
crc/baselines/contrastive_crl/src/nonlinearities.py - Model training
crc/baselines/contrastive_crl/src/training.py - Model evaluation
crc/baselines/contrastive_crl/src/evaluation.py - MCC metrics
crc/baselines/contrastive_crl/src/mcc.py
Shared runtime
- Shared architectures
crc/shared/architectures.py - Torch utilities
crc/utils/torch_utils.py - Task and split utilities
crc/utils/utils.py
Other components
- Experiment runner
- Causal Chamber data
Connections 26
- Experiment runner → Contrastive experiment: runs
README.md - Experiment runner → Multiview experiment: runs
README.md - Experiment runner → CITRIS experiment: runs
README.md - Experiment runner → Supervised fit: runs
- Causal Chamber data → Chamber datasets: supplies data
crc/shared/torch_datasets.py - Contrastive experiment → Data selection: selects data
crc/contrastive_crl_experiment.py - Contrastive experiment → Contrastive models: builds model
crc/contrastive_crl_experiment.py - Contrastive experiment → Model training: trains model
crc/contrastive_crl_experiment.py - Contrastive experiment → Model evaluation: evaluates model
crc/contrastive_crl_experiment.py - Data selection → Contrastive data: creates data
crc/baselines/contrastive_crl/src/utils.py - Data selection → Chamber datasets: loads chamber data
crc/baselines/contrastive_crl/src/utils.py - Contrastive data → Nonlinearities: uses transforms
crc/baselines/contrastive_crl/src/data_generation.py - Contrastive data → Contrastive models: uses model components
crc/baselines/contrastive_crl/src/data_generation.py - Contrastive models → Nonlinearities: uses nonlinearities
crc/baselines/contrastive_crl/src/models.py - Model training → Model evaluation: evaluates validation
crc/baselines/contrastive_crl/src/training.py - Model evaluation → MCC metrics: computes MCC
crc/baselines/contrastive_crl/src/evaluation.py - Baseline runner → Model training: trains model
crc/baselines/contrastive_crl/src/run_experiments.py - Baseline runner → Model evaluation: evaluates model
crc/baselines/contrastive_crl/src/run_experiments.py - Baseline runner → Contrastive models: builds model
crc/baselines/contrastive_crl/src/run_experiments.py - Baseline runner → Contrastive data: generates data
crc/baselines/contrastive_crl/src/run_experiments.py - Multiview experiment → Shared architectures: uses architectures
crc/multiview_experiment.py - Multiview experiment → Chamber datasets: prepares chamber data
- CITRIS experiment → Causal Chamber data: uses benchmark data
README.md - Supervised fit → Shared architectures: uses architectures
crc/supervised_fit_experiment.py - Contrastive experiment → Torch utilities: selects device
crc/contrastive_crl_experiment.py - Multiview experiment → Task and split utilities: uses task utilities
crc/multiview_experiment.py