Skip to content

PIPELINE_COMMANDS.md

Latest commit

 

History

History
279 lines (219 loc) · 8.03 KB

PIPELINE_COMMANDS.md

File metadata and controls

279 lines (219 loc) · 8.03 KB

Multi-Condition LOOCV Pipeline - Complete Commands

Overview

This document contains all commands for running the multi-condition LOOCV regression pipeline with baseline weight comparison and prediction visualization.

Configuration Used:

  • Data: 6 odors (no citral)
  • Features: 13 receptors (intersection mode, threshold=0.05)
  • Model: ElasticNet (l1_ratio=0.3)
  • LOOCV: Leave-one-odor-out (6 folds)

1. Regenerate Baseline Weights (One-time Setup)

This creates the baseline weights from the control condition (opto_AIR).

# Option A: Using the provided script
bash scripts/regenerate_baseline.sh

# Option B: Manual command sequence
python scripts/run_multicond_loocv.py \
  --csv /tmp/reaction_rates_no_citral.csv \
  --control-row opto_AIR \
  --conditions opto_AIR,opto_EB,opto_hex,opto_benz_1,opto_3-oct \
  --model elasticnet \
  --feature-set intersection \
  --activation-threshold 0.05 \
  --l1-ratio 0.3 \
  --outdir out/multicond_loocv_baseline \
  --seed 0

# Then extract baseline weights:
python << 'EOF'
import pandas as pd
df = pd.read_csv('out/multicond_loocv_baseline/weights_mean_opto_AIR.csv')
df = df.rename(columns={'mean_w': 'baseline_w'})
df[['feature', 'baseline_w']].to_csv('/tmp/baseline_weights_intersection.csv', index=False)
print(f'✓ Baseline weights saved to /tmp/baseline_weights_intersection.csv')
EOF

Output:

  • Baseline weights: /tmp/baseline_weights_intersection.csv
  • LOOCV results: out/multicond_loocv_baseline/

2. Run Full LOOCV with Baseline Weights and Plots

Runs LOOCV with visualizations showing baseline vs delta weights.

python scripts/run_multicond_loocv.py \
  --csv /tmp/reaction_rates_no_citral.csv \
  --control-row opto_AIR \
  --conditions opto_AIR,opto_EB,opto_hex,opto_benz_1,opto_3-oct \
  --model elasticnet \
  --feature-set intersection \
  --activation-threshold 0.05 \
  --l1-ratio 0.3 \
  --plot \
  --plot-top-n 13 \
  --plot-baseline-weights /tmp/baseline_weights_intersection.csv \
  --plot-comparison \
  --outdir out/multicond_loocv_best \
  --seed 0

Output:

  • LOOCV results: out/multicond_loocv_best/
  • Plots (baseline in purple): out/multicond_loocv_best/plots/*.png
    • weights_deltaper_*.png (6 odor-specific plots with baseline vs delta weights)
    • weights_all_conditions.png (cross-condition comparison)

3. Make Predictions with Averaged Weights

Uses the mean weights from LOOCV to predict ΔPER and compare to true values.

python scripts/predict_with_avg_weights.py \
  --loocv-dir out/multicond_loocv_best \
  --csv /tmp/reaction_rates_no_citral.csv \
  --control-row opto_AIR \
  --conditions opto_EB,opto_hex,opto_benz_1,opto_3-oct \
  --feature-set intersection \
  --activation-threshold 0.05 \
  --outdir out/prediction_plots_best

Output:

  • Prediction plots: out/prediction_plots_best/
    • predictions_vs_true.png (bar plots per condition)
    • predictions_scatter.png (scatter: true vs predicted)
    • weights_comparison.png (averaged weights across conditions)
  • Predictions CSV: out/prediction_plots_best/predictions.csv

CSV Columns:

  • condition: trained condition name
  • odor: odor name
  • true_delta_per: true ΔPER value
  • predicted_delta_per: predicted ΔPER value
  • true_centered: true ΔPER (mean-centered)
  • predicted_centered: predicted ΔPER (mean-centered)

4. Quick Full Pipeline (All Steps)

Run everything in sequence:

#!/bin/bash

# Step 1: Regenerate baseline
bash scripts/regenerate_baseline.sh

# Step 2: Run LOOCV with plots
python scripts/run_multicond_loocv.py \
  --csv /tmp/reaction_rates_no_citral.csv \
  --control-row opto_AIR \
  --conditions opto_AIR,opto_EB,opto_hex,opto_benz_1,opto_3-oct \
  --model elasticnet \
  --feature-set intersection \
  --activation-threshold 0.05 \
  --l1-ratio 0.3 \
  --plot --plot-top-n 13 \
  --plot-baseline-weights /tmp/baseline_weights_intersection.csv \
  --plot-comparison \
  --outdir out/multicond_loocv_best \
  --seed 0

# Step 3: Make predictions
python scripts/predict_with_avg_weights.py \
  --loocv-dir out/multicond_loocv_best \
  --csv /tmp/reaction_rates_no_citral.csv \
  --control-row opto_AIR \
  --conditions opto_EB,opto_hex,opto_benz_1,opto_3-oct \
  --feature-set intersection \
  --activation-threshold 0.05 \
  --outdir out/prediction_plots_best

echo "✓ Complete!"
echo ""
echo "Results:"
echo "  LOOCV plots: out/multicond_loocv_best/plots/"
echo "  Predictions: out/prediction_plots_best/"

5. View Results

# List all plots
ls -lh out/multicond_loocv_best/plots/
ls -lh out/prediction_plots_best/

# View predictions CSV
head -20 out/prediction_plots_best/predictions.csv

# View baseline weights
head -10 /tmp/baseline_weights_intersection.csv

Key Parameters Explained

Parameter Value Reason
--csv /tmp/reaction_rates_no_citral.csv 6 odors (citral removed)
--feature-set intersection Only receptors active in all 6 odors
--activation-threshold 0.05 Only responses > 0.05 count as "active"
--model elasticnet Balanced L1/L2 regularization
--l1-ratio 0.3 30% L1 (Lasso), 70% L2 (Ridge)
--plot-top-n 13 Show all 13 intersection receptors

Feature Set Details

Current Configuration:

  • Total receptors in dataset: 60
  • Receptors with any non-zero response: 57 (no_blanks)
  • Receptors active in ALL 6 odors: 13 (intersection)

The 13 Intersection Receptors:

Or19a, Or22a, Or2a, Or35a, Or47b, Or67b, Or7a, Or83c,
Or85b, Or98a, ac1, ac2, ac3_noOr35a

Expected Performance

With 13 receptors (intersection) and 6 odors (no citral):

Condition R² MSE Notes
opto_EB 0.40 0.009 Good prediction
opto_hex 0.44 0.032 Best prediction
opto_benz_1 0.30 0.037 Moderate
opto_3-oct 0.27 0.004 Moderate

Overall R²: ~0.35 (averaged across 4 trained conditions)

Troubleshooting

Baseline weights are all zero

Problem: Using baseline from opto_AIR control when it has poor fit (R² < 0)

Solution: Ensure you extracted baseline AFTER fitting the LOOCV

python scripts/run_multicond_loocv.py ... --outdir out/multicond_loocv_baseline
# Then extract from that directory's weights_mean_opto_AIR.csv

File not found errors

Problem: Script looking for weights_mean_opto_3_oct.csv but file is weights_mean_opto_3-oct.csv

Solution: Use consistent naming - the script handles opto_3-oct → opto_3_oct conversion

No citral CSV not found

Problem: /tmp/reaction_rates_no_citral.csv doesn't exist

Solution: Create it:

python << 'EOF'
import pandas as pd
df = pd.read_csv('/home/ramanlab/Documents/cole/Results/Opto/Reaction_Predictions/reaction_rates_summary_unordered.csv')
df = df.drop(columns=['Citral'])
df.to_csv('/tmp/reaction_rates_no_citral.csv', index=False)
EOF

File Structure

out/
├── multicond_loocv_baseline/      (Initial baseline run)
│   ├── conditions_overview.csv
│   ├── weights_mean_opto_AIR.csv  ← Extract baseline from here
│   ├── weights_mean_opto_EB.csv
│   └── ...
├── multicond_loocv_best/          (Main LOOCV run with plots)
│   ├── conditions_overview.csv
│   ├── plots/
│   │   ├── weights_deltaper_*.png
│   │   └── weights_all_conditions.png
│   ├── weights_mean_*.csv
│   └── ...
└── prediction_plots_best/         (Predictions from averaged weights)
    ├── predictions_vs_true.png
    ├── predictions_scatter.png
    ├── weights_comparison.png
    └── predictions.csv

/tmp/
└── baseline_weights_intersection.csv  (Baseline weights for plotting)

Notes

  • All commands assume you're in the door-python-toolkit root directory
  • The --seed 0 ensures reproducible results
  • Baseline weights are computed from control condition only (opto_AIR raw PER)
  • Delta weights show change from baseline in each trained condition
  • Predictions use averaged LOOCV weights (mean across 6 folds)