get_decay_configs.py

import numpy as np
import pickle
import json
import os
import matplotlib.pyplot as plt

from city_conf import city_mappings

WEIGHT_AT_UPPER_THRESHOLD = 0.1


def derive_exponential_decay_params(dists):
    """Calibrate decay to put weight 0.1 at 90th percentile of road distance from centroid,
    and 1.0 at 10th percentile"""
    lower_threshold = np.percentile(dists, 10)
    upper_threshold = np.minimum(np.percentile(dists, 90), 15)
    decay_coef = np.log(WEIGHT_AT_UPPER_THRESHOLD) / (upper_threshold - lower_threshold)

    return {
        "lower_threshold": round(lower_threshold, 1),
        "upper_threshold": round(upper_threshold, 1),
        "decay_coef": round(decay_coef, 2)
    }


for country_map in city_mappings:
    for city in city_mappings[country_map]:
        city_name = list(city.keys())[0]
        if not (os.path.exists(f"results/{city_name}_distances.pkl") and os.path.exists(
                f"results/{city_name}_decay_conf.json")):
            print(f"{city_name} - working")
            with open(f"results/{city_name}_distances.pkl", "rb") as f:
                dists = pickle.load(f)

            decay_conf = derive_exponential_decay_params(dists)
            with open(f"results/{city_name}_decay_conf.json", "w") as f:
                json.dump(decay_conf, f)

            plt.hist(dists, bins=100)
            plt.axvline(x=decay_conf["lower_threshold"], color="red")
            plt.axvline(x=decay_conf["upper_threshold"], color="red")
            plt.title(city_name)
            plt.savefig(f"results/{city_name}_distance_plot.png")
            plt.close()
        else:
            print(f"{city_name} - skipped")