Skip to content

Latest commit

 

History

History
117 lines (89 loc) · 3.56 KB

README.md

File metadata and controls

117 lines (89 loc) · 3.56 KB

Home / Batch

Batch System

Multi-run consensus and parameter sweeps for Skibidy. Self-contained: all scripts, configs, and results live here.

Multi-run consensus

Run N simulations with the same config but different random seeds, compute mean/std, and validate against literature:

python3 batch/batch.py                         # 20 runs, default config
python3 batch/batch.py -n 5                    # 5 runs (faster)
python3 batch/batch.py -n 10 --study wound     # wound study, 10 runs
python3 batch/batch.py --skin aged --study wound --validate

Output: batch/results/consensus_<skin>_<study>_<timestamp>/

  • raw/run_NNN.csv: individual run metrics
  • metrics.csv: mean values (validation-compatible)
  • consensus.csv: mean + std for each column

Parameter sweeps

Sweep one or two parameters across a range of values with replicates at each point:

python3 batch/sweep.py batch/configs/cytokine_rate.toml       # single-param sweep
python3 batch/sweep.py batch/configs/collagen_vs_scar.toml     # two-param heatmap
python3 batch/sweep.py batch/configs/cytokine_rate.toml -n 3   # override replicates
python3 batch/sweep.py batch/configs/diabetic_factors.toml --analyze

Output: batch/results/<name>_<timestamp>/

  • config.toml: copy of sweep config
  • raw/ptNNN_val*_runNNN.csv: individual run metrics
  • summary.csv: param values + outcome mean/std at each point
  • sweep.png: auto-generated plot (single-param with error bars)
  • heatmap.png: auto-generated plot (two-param colored grid)

Post-hoc analysis

Analyze any results directory after the fact:

python3 batch/analyze.py batch/results/cytokine_rate_20260225_143000/
python3 batch/analyze.py batch/results/consensus_default_wound_20260225/ --validate

Sweep config format

Configs are TOML files in batch/configs/. Single parameter:

[sweep]
name = "cytokine_rate"
param = "skin.immune.cytokine_rate"
values = [0.0005, 0.001, 0.0015, 0.002, 0.003]
runs_per_value = 5
study = "wound"           # optional: apply studies/wound/preset.toml

[outcomes]
primary = "wound_closure_pct"
secondary = ["mean_infl_wound", "n_macrophages"]
measure = "final"         # final, peak, auc, time_to_90

Two parameters (produces NxM grid):

[sweep]
name = "collagen_vs_scar"
runs_per_value = 3

[[sweep.params]]
param = "skin.fibroblast.collagen_deposition_rate"
values = [0.004, 0.008, 0.012, 0.016]

[[sweep.params]]
param = "skin.scar.collagen_threshold"
values = [0.05, 0.1, 0.2, 0.5]

[outcomes]
primary = "scar_magnitude"
measure = "final"

Included configs

Config Type What it tests
cytokine_rate.toml 1D Inflammation production rate vs healing outcome
resolution_rate.toml 1D M2 resolution speed vs inflammation duration
collagen_vs_scar.toml 2D Collagen deposition vs scar threshold (heatmap)
diabetic_factors.toml 1D M1 duration factor vs diabetic closure delay

Outcome measures

Measure Description
final Last value in the timeseries
peak Maximum value
auc Area under curve (trapezoidal, using time_h)
time_to_N First time value exceeds N (e.g. time_to_90 for 90% closure)

Directory structure

batch/
  batch.py           Multi-run consensus runner
  sweep.py           Parameter sweep runner
  analyze.py         Post-hoc analysis and plotting
  lib.py             Shared helpers (config, simulation, CSV, outcomes)
  configs/           Sweep config files
  results/           Output (gitignored)
  README.md