RLWM Trauma Analysis

Reinforcement Learning Working Memory (RLWM) task analysis pipeline — fitting computational models (Q-learning, WM-RL) to behavioral data and relating parameters to trauma exposure (LEC-5) and PTSD symptoms (IES-R). Based on Senta et al. (2025).

Setup

conda env create -f environment.yml
conda activate ds_env

Pipeline

Run scripts sequentially (01-16). Each stage depends on the previous.

Data Processing (01-04)

python scripts/01_parse_raw_data.py          # Parse raw jsPsych JSON
python scripts/02_create_collated_csv.py      # Collate participants (wide)
python scripts/03_create_task_trials_csv.py   # Trial-level data (long)
python scripts/04_create_summary_csv.py       # Summary metrics

Behavioral Analysis (05-08)

python scripts/05_summarize_behavioral_data.py     # Descriptive stats
python scripts/06_visualize_task_performance.py     # Learning curves, set-size effects
python scripts/07_analyze_trauma_groups.py          # Trauma grouping + clustering
python scripts/08_run_statistical_analyses.py       # ANOVAs, descriptive tables

Simulations & Validation (09-11)

python scripts/09_generate_synthetic_data.py       # Synthetic data generation
python scripts/10_run_parameter_sweep.py            # Parameter space exploration
python scripts/11_run_model_recovery.py             # Parameter & model recovery

Model Fitting (12-14)

python scripts/12_fit_mle.py --model qlearning      # Q-learning (M1)
python scripts/12_fit_mle.py --model wmrl            # WM-RL (M2)
python scripts/12_fit_mle.py --model wmrl_m3         # WM-RL + perseveration (M3)
python scripts/14_compare_models.py                  # AIC/BIC model comparison

For parallel or GPU-accelerated fitting:

python scripts/12_fit_mle.py --model wmrl_m3 --n-jobs 16     # Multi-core
python scripts/12_fit_mle.py --model wmrl_m3 --use-gpu       # GPU (rlwm_gpu env)

Results Analysis (15-16)

python scripts/15_analyze_mle_by_trauma.py --model all        # Parameter-trauma groups
python scripts/16_regress_parameters_on_scales.py --model all  # Continuous regressions

Documentation

Document	Contents
ANALYSIS_PIPELINE.md	Full pipeline walkthrough (stages 1-5), fitting options, IC reference
MODEL_REFERENCE.md	Model math: Q-learning, WM-RL, parameter bounds, priors
TASK_AND_ENVIRONMENT.md	Task structure, environment API, block/trial layout
PARTICIPANT_EXCLUSIONS.md	Exclusion criteria, final N=48

Data

Input

Raw jsPsych CSV files go in data/. One file per participant session.

Key Outputs

File	Description
output/task_trials_long.csv	Main task trials (long format, for fitting)
output/task_trials_long_all.csv	All trials including practice blocks
output/collated_participant_data.csv	All participant data (wide format)
output/summary_participant_metrics.csv	Per-participant summary metrics
output/mle/	MLE fitting results per model
output/model_comparison/	Model comparison tables
output/trauma_groups/	Trauma group assignments
figures/	All generated visualizations

Models

Model	Parameters	Description
Q-learning (M1)	alpha_pos, alpha_neg, epsilon	Asymmetric RL
WM-RL (M2)	alpha_pos, alpha_neg, phi, rho, K, epsilon	RL + working memory
WM-RL+P (M3)	+ stick	M2 + perseveration

Fixed beta=50 during learning (for identifiability). Epsilon captures random responding. See MODEL_REFERENCE.md for full math.

Tests

python -m pytest scripts/fitting/tests/ -v

Configuration

python config.py  # Print current settings