terrassumptions/server/lib.py

import json
import math
import random

import requests
import haversine

# Google API key, with access to Street View Static API
google_api_key = "AIzaSyAqjCYR6Szph0X0H_iD6O1HenFhL9jySOo"
metadata_url = "https://maps.googleapis.com/maps/api/streetview/metadata"
mapcrunch_url = "http://www.mapcrunch.com/_r/"
rsv_url = "https://randomstreetview.com/data"
urban_centers = []
with open("./urban-centers.csv") as infile:
    for line in infile:
        lat, lng = line.split(",")
        urban_centers.append((float(lat.strip()), float(lng.strip())))


def point_has_streetview(lat, lng):
    """
    Returns True if the streetview metadata endpoint says a given point has
    data available, and False otherwise.

    This function calls the streetview metadata endpoint - there is no quota consumed.
    """
    params = {
        "key": google_api_key,
        "location": f"{lat},{lng}",
    }
    js = requests.get(metadata_url, params=params).json()
    return js["status"] == "OK"


def generate_coord(max_retries=100, only_america=False):
    """
    Returns (latitude, longitude) of usable coord (where google has data).
    This function will attempt at most max_retries calls to map crunch to fetch
    candidate points, and will exit as soon as a suitable candidate is found.
    If no suitable candidate is found in this allotted number of retries, None is
    returned.

    This function calls the streetview metadata endpoint - there is no quota consumed.
    """
    mc_url = mapcrunch_url + ("?c=21" if only_america else "")
    for _ in range(max_retries):
        points_res = requests.get(mc_url).text
        points_js = json.loads(points_res.strip("while(1); "))
        if "c=" not in mc_url:
            mc_url += f"?c={points_js['country']}" # lock to the first country randomed
        for lat, lng in points_js["points"]:
            if point_has_streetview(lat, lng):
                return (lat, lng)


def call_random_street_view(only_america=False):
    """
    Returns an array of (some number of) tuples, each being (latitude, longitude).
    All points will be valid streetview coordinates. There is no guarantee as to the
    length of this array (it may be empty), but it will never be None.

    This function calls the streetview metadata endpoint - there is no quota consumed.
    """
    rsv_js = requests.post(rsv_url, data={"country": "us" if only_america else "all"}).json()
    if not rsv_js["success"]:
        return []
    
    return [
        (point["lat"], point["lng"])
        for point in rsv_js["locations"]
        if point_has_streetview(point["lat"], point["lng"])
    ]


def random_street_view_generator(only_america=False):
    """
    Returns a generator which will lazily use call_random_street_view to generate new
    street view points.

    The returned generator calls the streetview metadata endpoint - there is no quota consumed.
    """
    points = []
    while True:
        if len(points) == 0: 
            points = call_random_street_view(only_america=only_america)
        else:
            yield points.pop()


def urban_coord(max_retries=100, max_dist_km=25):
    """
    Returns (latitude, longitude) of usable coord (where google has data) that is near
    a known urban center. Points will be at most max_dist_km kilometers away. This function will
    check at most max_retries points, and if none have street view data, will return None.
    Otherwise, it will exit as soon as suitable point is found.

    This function calls the streetview metadata endpoint - there is no quota consumed.
    """

    for _ in range(max_retries):
        (city_lat, city_lng) = random.choice(urban_centers)
        dist_km = random.random() * max_dist_km
        angle_rad = random.random() * 2 * math.pi
        # logic adapted from https://stackoverflow.com/a/7835325
        city_lat_rad = math.radians(city_lat)
        city_lng_rad = math.radians(city_lng)
        pt_lat_rad = math.asin(math.sin(city_lat_rad) * math.cos(dist_km / mean_earth_radius_km) + math.cos(city_lat_rad) * math.sin(dist_km / mean_earth_radius_km) * math.cos(angle_rad))
        pt_lng_rad = city_lng_rad + math.atan2(math.sin(angle_rad) * math.sin(dist_km / mean_earth_radius_km) * math.cos(city_lat_rad), math.cos(dist_km / mean_earth_radius_km) - math.sin(city_lat_rad) * math.sin(pt_lat_rad))
        pt_lat = math.degrees(pt_lat_rad)
        pt_lng = math.degrees(pt_lng_rad)
        if point_has_streetview(pt_lat, pt_lng):
            return (pt_lat, pt_lng)


mean_earth_radius_km = (6378 + 6357) / 2

# if you're more than 1/4 of the Earth's circumfrence away, you get 0
max_dist_km = (math.pi * mean_earth_radius_km) / 2 # this is about 10,000 km

# if you're within 1/16 of the Earth's circumfrence away, you get at least 1000 points
quarter_of_max_km = max_dist_km / 4 # this is about 2,500 km

# https://www.wolframalpha.com/input/?i=sqrt%28%28%28land+mass+of+earth%29+%2F+7%29%29+%2F+pi%29+in+kilometers
# this is the average "radius" of a continent
# within this radius, you get at least 2000 points
avg_continental_rad_km = 1468.0

# somewhat arbitrarily, if you're within 1000 km, you get at least 3000 points
one_thousand = 1000.0

# https://www.wolframalpha.com/input/?i=sqrt%28%28%28land+mass+of+earth%29+%2F+%28number+of+countries+on+earth%29%29+%2F+pi%29+in+kilometers
# this is the average "radius" of a country
# within this radius, you get at least 4000 points
avg_country_rad_km = 479.7

# if you're within 150m, you get a perfect score of 5000
min_dist_km = 0.15  


def score_within(raw_dist, min_dist, max_dist):
    """
    Gives a score between 0 and 1000, with 1000 for the min_dist and 0 for the max_dist
    """
    # scale the distance down to [0.0, 1.0], then multiply it by 2 for easing
    pd2 = 2 * (raw_dist - min_dist) / (max_dist - min_dist)
    # perform a quadratic ease-in-out on pd2
    r = (pd2 ** 2) / 2 if pd2 < 1 else 1 - (((2 - pd2) ** 2) / 2)
    # use this to ease between 1000 and 0
    return int(1000 * (1 - r))


def score(target, guess):
    """
    Takes in two (latitude, longitude) pairs and produces an int score.
    Score is in the (inclusive) range [0, 5000]
    Higher scores are closer.

    Returns (score, distance in km)
    """
    dist_km = haversine.haversine(target, guess)

    if dist_km <= min_dist_km:
        point_score = 5000
    elif dist_km <= avg_country_rad_km:
        point_score = 4000 + score_within(dist_km, min_dist_km, avg_country_rad_km)
    elif dist_km <= one_thousand:
        point_score = 3000 + score_within(dist_km, avg_country_rad_km, one_thousand)
    elif dist_km <= avg_continental_rad_km:
        point_score = 2000 + score_within(dist_km, one_thousand, avg_continental_rad_km)
    elif dist_km <= quarter_of_max_km:
        point_score = 1000 + score_within(dist_km, avg_continental_rad_km, quarter_of_max_km)
    elif dist_km <= max_dist_km:
        point_score = score_within(dist_km, quarter_of_max_km, max_dist_km)
    else: # dist_km > max_dist_km
        point_score = 0

    return point_score, dist_km