testmain.py

import math
from pprint import pprint
import random
import scipy.spatial.distance
import numpy as np

pred = [.7,.1,.1,.1]
true = [.5,.2,.2,.1]

def calc_prob(history_list):
    prob = [0,0,0,0]
    for item in history_list:
        prob[item] += 1
    for i in range(len(prob)):
        prob[i] = round(prob[i]/len(history_list), 3)
    return prob


def model_distance(accept, test):
    return scipy.spatial.distance.cosine(accept, test)


def emulate_room(room):
    pass


def calc_trajectory(history):
    hypothesis = []
    hypothesis_dist = []
    hypothesis_split = []
    max_range = 10
    half_length = math.floor(len(history)/2)
    variance = (len(history) / 5)

    # split_point = half_length + random.randint(-variance, variance)
    split_points = []
    for i in range(max_range):
        split_points.append(half_length + math.floor((variance / max_range) * (i - math.ceil(max_range / 2))))

    split_points.sort()
    for split_point in split_points:

        new = history[split_point:]
        old = history[:split_point]

        # given a random split between the old and the new, figure out where the new should be?

        # calculate probability of new. thats a hypothesis.
        new_prob = np.array(calc_prob(new)) / np.linalg.norm(new)
        old_prob = np.array(calc_prob(old)) / np.linalg.norm(old)
        expected_prob = []

        # find angle
        np.arccos(np.clip(np.dot(new_prob, old_prob), -1.0, 1.0))

        # rotate by that angle

        # ???

        # profit!

        hypothesis.append(new_prob)
        hypothesis_split.append(split_point)
        hypothesis_dist.append(round(model_distance(true, new_prob), 3))

    for i in range(len(hypothesis)):
        print(hypothesis_split[i], ":", hypothesis_dist[i], ",", hypothesis[i])

    pass


history_pre = random.choices([0, 1, 2, 3], weights=true, k=50)
history_post= random.choices([0, 1, 2, 3], weights=pred, k=50)

history_comb = history_pre + history_post

calc_trajectory(history_comb)
# print(calc_prob(history_pre))
# print(calc_prob(history_post))
# for i in range(0, 51, 10):
#     calc = calc_prob(history_comb[i:i+50])
#     distance = model_distance(true, calc)
    # print(i, calc, distance)