Tutorials#
Interactive marimo notebooks under tutorials/ in
the repository. They are not rendered here; run one with
pip install "chem-tactics[tutorials]"
marimo edit tutorials/<name>.py # or: marimo run ... for app mode
Working with the package#
thompson_sampling_tutorial.py — pick strategies and warmups, run
them on the bundled thrombin library, and compare recovery with
TS_Benchmarks. Blocks 1–3, 6.
library_enumeration_tutorial.py — single-step, multi-step and
alternative-SMARTS pipelines, full enumeration, then a search.
Block 1, with a taste of 3.
reaction_config_builder.py — build a ReactionConfig in a form,
validate it against your reagent files, see the failures. Block 1.
custom_evaluator_tester.py — paste a scoring function, check it
compiles and behaves, run a short search with it. Block 2.
diagnostic_benchmark_plots.py — the mechanism plots
(plot_rws_diagnostic, plot_ttts_diagnostic, the layered
panels) over the published diagnostic benchmark. Blocks 5–6.
Needs the benchmark parquet under outputs/.
interactive_sar_explorer.py — hover a point, see the structure;
per-component oracle GMIC over the ground-truth scores. Needs
data/scores.
Manuscript figures#
Reproduce the paper’s figures from the benchmark output (not the package
API). Each expects the benchmark parquet files under outputs/.
manuscript_plots_ROCS.py— ROCS libraries: aggregate top-N recovery, per-library breakdowns, Tukey HSD, budget sensitivity.manuscript_plots_docking.py— the eight docking libraries: method × library heatmap, significance, budget sensitivity.manuscript_sar_plots.py— reagent score landscapes from the brute-force ground truth.