How jerenemy/polydiff works
Polydiff generates fixed- or variable-size 2D polygon datasets, trains MLP/GAT/GCN DDPM denoisers, samples new polygons, and evaluates results with diagnostics. A study runner orchestrates repeatable training and sampling sweeps; pocket-conditioning and binding-affinity surrogate experiments add guidance-oriented capabilities. The graph includes the documented workflow and source-index-supported calls. Some study, guidance, and integration details are represented only at subsystem level because their implementation excerpts were incomplete.
Main entry points: run.py Study runner · train.py Diffusion training · runtime.py Sampling runtime · study.py Pocket conditioning
Components 25
Polygon data
- Polygon generator
polydiff/data/gen_polygons.py - Polygon datasets
polydiff/data/polygon_dataset.py - Polygon diagnostics
polydiff/data/diagnostics.py - Polygon plotting
polydiff/data/plot_polygons.py
Diffusion models
- DDPM diffusion
polydiff/models/diffusion.py - GAT denoiser
polydiff/models/gat.py - GCN denoiser
polydiff/models/gcn.py - Guidance models
polydiff/models/guidance_models.py
Training and sampling
- Diffusion training
polydiff/training/train.py - Guidance training
polydiff/training/train_guidance_model.py - Run organization
polydiff/runs.py - Sampling entrypoint
polydiff/sampling/sample.py - Sampling runtime
polydiff/sampling/runtime.py - Sampling guidance
polydiff/sampling/guidance.py - Restoration metrics
polydiff/restoration.py
Studies and guidance
- Study runner
polydiff/studies/run.py - Study runtime
polydiff/studies/runtime.py - Pocket conditioning
polydiff/pocket_conditioning/study.py - Pocket pair data
polydiff/pocket_conditioning/data.py - Pocket geometry
polydiff/pocket_conditioning/geometry.py - Pocket surrogate
polydiff/pocket_conditioning/surrogate.py
Affinity surrogate
- Affinity data
predict_binding_affinity/surrogate_data.py - Affinity models
predict_binding_affinity/surrogate_models.py - Surrogate training
predict_binding_affinity/train_surrogate_model.py
Other components
- Researcher
Connections 29
- Researcher → Polygon generator: generates data
- Polygon generator → Polygon datasets: writes formats
polydiff/data/gen_polygons.py - Researcher → Diffusion training: starts training
- Diffusion training → Polygon datasets: loads data
polydiff/training/train.py - Diffusion training → DDPM diffusion: trains model
polydiff/training/train.py - DDPM diffusion → GAT denoiser: builds denoiser
polydiff/models/diffusion.py - DDPM diffusion → GCN denoiser: builds denoiser
polydiff/models/diffusion.py - Guidance training → Polygon datasets: loads data
polydiff/training/train_guidance_model.py - Guidance training → Guidance models: trains guidance
polydiff/training/train_guidance_model.py - Diffusion training → Run organization: creates run
polydiff/training/train.py - Researcher → Sampling entrypoint: starts sampling
- Sampling entrypoint → Sampling runtime: runs sampling
polydiff/sampling/sample.py - Sampling runtime → DDPM diffusion: rebuilds model
polydiff/sampling/runtime.py - Sampling entrypoint → Sampling guidance: configures guidance
polydiff/sampling/sample.py - Sampling runtime → Polygon diagnostics: computes diagnostics
polydiff/sampling/runtime.py - Sampling runtime → Restoration metrics: tracks restoration
polydiff/sampling/runtime.py - Researcher → Polygon plotting: plots polygons
- Researcher → Study runner: runs study
- Study runner → Diffusion training: dispatches training
polydiff/studies/run.py - Study runner → Guidance training: dispatches training
polydiff/studies/run.py - Study runner → Sampling entrypoint: dispatches sampling
polydiff/studies/run.py - Study runner → Study runtime: uses runtime
polydiff/studies/run.py - Researcher → Pocket conditioning: runs conditioning
- Pocket conditioning → Pocket pair data: prepares pairs
polydiff/pocket_conditioning/study.py - Pocket conditioning → Pocket geometry: uses geometry
polydiff/pocket_conditioning/study.py - Pocket conditioning → Pocket surrogate: uses surrogate
polydiff/pocket_conditioning/study.py - Surrogate training → Affinity data: loads data
predict_binding_affinity/train_surrogate_model.py - Surrogate training → Affinity models: trains model
predict_binding_affinity/train_surrogate_model.py - Affinity models → Affinity data: uses data
predict_binding_affinity/surrogate_models.py