Workspace Issues
Results
Metric cell shading (per column, best value first):
1st
2nd
3rd
reference row
μ replica mean (multi-replica cells only)
σ std dev in those cells
Δ vs reference (column)
+ better than reference
− worse than reference
~ |Δ| < σ (within replica noise)
≈ learned SF classifier (lr / rf / xgb / …)
Metric cell shading (per column, best value first):
1st
2nd
3rd
reference row
μ replica mean (multi-replica cells only)
σ std dev in those cells
Δ vs reference (column)
+ better than reference
− worse than reference
~ |Δ| < σ (within replica noise)
≈ learned SF classifier (lr / rf / xgb / …)
Cross-validation
Details
Design ablation
Compose a feature-policy YAML and training command without leaving the dashboard.Feature policy
Start from template
Identity
Available features
Uses raw_prepare/raw_pdbbind.csv and raw_prepare/raw_dudez.csv under the served Workbench root. Metadata and target columns are stripped automatically.
Feature filter is case-insensitive and only narrows the browser — it does not affect policy matching.
Use wildcards or checkboxes, then apply to include_features (exact names) or include_patterns / exclude fields. Patterns use fnmatch (*, ?).
Include / exclude rules
Run settings
Paths used when generating the training command. The dashboard stays read-only: save YAML locally, then run the command in your shell.
Preview & plan
VS design
Discover inputs
Best-effort scan for receptor/ligand/box candidates under the served root (or a subdirectory) by filename/extension — not a fixed layout. Pick a candidate below, or type any path directly in Run settings.
Run settings
Preview & plan
Discover campaign samples
Best-effort scan for an input/{sample}/{receptor,ligand,box} layout —
one subdirectory per sample, matching the bundled example Snakefiles. Missing that
layout, edit the manifest below by hand.
Manifest
One JSON object per sample: sample, row_kind
("vs" or "pipeline"), receptor,
ligand, box, engines (list), optional
rescoring_engines (list). Rows may mix vs and
pipeline.
Preview & plan
Job token
Job-execute endpoints (/api/jobs*) require a bearer token. It is auto-generated on first
ocdocker workbench serve run at ~/.config/ocdocker/workbench_token, or set via
OCDOCKER_WORKBENCH_TOKEN. Paste it below to launch or cancel jobs from this browser; it is
kept in this browser's local storage only and never sent anywhere but this API.