bootstrap_SCV.py

import argparse
import numpy as np
from utils.evaluation import eval_SCV, bootstrap_SCV


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('-a', '--AV', type=int, choices=[0, 1, 2], 
                        default=0, help='The AV model under test')
    parser.add_argument('-b', '--num_bootstrap', type=int, 
                        default=1000, help='Number of bootstrap')
    parser.add_argument('-s', '--num_separation', type=int, 
                        default=100, help='Number of separation')
    parser.add_argument('-t', '--RHW_threshold', type=float, 
                        default=0.3, help='The threshold for relative half-width')
    args = parser.parse_args()
    AV = args.AV
    num_bootstrap = args.num_bootstrap
    num_separation = args.num_separation
    RHW_threshold = args.RHW_threshold
    print(f'Bootstrapping SCV, AV = {AV}, num_bootstrap = {num_bootstrap}')

    crash_NADE = np.load(f'results/crash_NADE_AV_{AV}.npy').flatten()
    control_step = np.load(f'results/control_step_AV_{AV}.npy').flatten()
    control_variates = np.load(f'results/control_variates_AV_{AV}.npy', allow_pickle=True)
    CV = {}
    for i in range(len(control_variates)):
        for j in range(len(control_variates[0])):
            CV[j + len(control_variates[0]) * i] = control_variates[i][j]
    
    crash_SCV_dict, rhw_SCV = eval_SCV(crash_NADE, control_step, CV, sep=num_separation)
    crash_SCV = np.zeros_like(crash_NADE)
    for l in list(crash_SCV_dict.keys()):
        idx = (control_step == l)
        crash_SCV[idx] = crash_SCV_dict[l]
    np.save(f"results/crash_SCV_AV_{AV}", crash_SCV)
    np.save(f"results/rhw_SCV_total_AV_{AV}", rhw_SCV)

    RNoT_SCV = bootstrap_SCV(crash_NADE, control_step, CV, 
                             n_bootstrap=num_bootstrap, 
                             sep=num_separation, 
                             RHW_threshold=RHW_threshold)
    np.save(f'results/RNoT_SCV_bootstrap_{num_bootstrap}_AV_{AV}', RNoT_SCV)