Running Simulations

CLI usage

The main entry point is run_policy.py:

python run_policy.py <policy_name> <repeat> <repeat_start> <output_dir> [options]

Positional arguments:

Argument	Description
`policy_name`	Policy to run (e.g., `null`, `high-risk`, `low-risk`, `age-first`)
`repeat`	Number of repetitions
`repeat_start`	Starting index for repetition numbering
`output_dir`	Base directory for output files

Simulation parameters (--):

Flag	Default	Description
`--T_max`	40000	Maximum simulation time (days)
`--p_length`	100	Episode length (days)
`--p_freeze`	2	Periods to freeze before updating scores
`--p_freeze_policy`	10	Periods before updating the policy
`--rel_off_probation`	2000	Off-probation term (days)
`--beta_arrival`	5	Arrival rate parameter
`--beta_initial`	5	Initial population size

Treatment parameters:

Flag	Default	Description
`--treatment_capacity`	80	Treatment slots per episode
`--treatment_effect`	0.5	Effect size (0-1), or `type-1` for heterogeneous
`--treatment_dosage`	default	`default` (1.0) or `type-1` for heterogeneous

Example:

python run_policy.py high-risk 10 0 results --T_max 60000 --treatment_capacity 100

Batch runs

Shell scripts in sbin/ generate and run batches of policy simulations:

# Generate commands for 10 repeats starting at index 0, output to results/
zsh sbin/run_all.zsh 10 0 results

# This creates results/cmd.sh which can be executed:
bash results/cmd.sh

Available batch scripts:

Script	Description
`run_all.zsh`	Standard run across policies
`run_all_params_cap.zsh`	Vary treatment capacity
`run_all_params-dosage.zsh`	Vary dosage
`run_all_params-dosage-type1.zsh`	Type-1 heterogeneous dosage
`run_all_params-less-people.zsh`	Smaller population
`run_all_params-more-people.zsh`	Larger population
`run_all_params-ofpl.zsh`	Vary off-probation length
`run_all_params-younger.zsh`	Younger age distribution

Environment variables

Variable	Default	Description
`DBG`	`0`	`1` for debug mode (more arrivals, faster iteration)
`TR_LOW`	`15.0`	Treatment effect lower bound
`TR_HIGH`	`0.0`	Treatment effect upper bound

Sensitivity analysis

Sensitivity analysis scripts live in sensanaly/:

python sensanaly/run_sensanaly_capeff.py   # capacity effect
python sensanaly/run_sensanaly_prscale.py  # probability scale