histogram.py

from sklearn.model_selection import train_test_split
import pickle
import os
import matplotlib.pyplot as plt

from mnist import load_mnist
from knn import learn_knn, knn_predict
import make_dir as md

OPT_K = 3


def distinguish_distances(distances, y_pred, y_test):  # correct or wrong
    correct_dis = []
    wrong_dis = []
    for i, (p, t) in enumerate(zip(y_pred, y_test)):
        if p == t:
            correct_dis.append(distances[i])
        else:
            wrong_dis.append(distances[i])
    return correct_dis, wrong_dis


def histogram_with_correct_wrong(correct_dis, wrong_dis, is_show=False):
    base_dir = "images"
    md.make_dir(base_dir)
    fname = f"{base_dir}/histogram_cw.png"
    longest = int(max(max(correct_dis), max(wrong_dis)))
    bins = [i for i in range(0, longest, 100)]
    _, axs = plt.subplots(2, 1, figsize=(8, 8))
    axs[0].set_title("Correct distances")
    axs[1].set_title("Wrong distances")
    axs[0].hist(correct_dis, bins=bins, color="blue")
    axs[1].hist(wrong_dis, bins=bins, color="red")
    plt.savefig(fname)
    if is_show:
        plt.show()


def make_knn_nearest_distacnes(knn, X_test):
    base_dir = "pickles"
    md.make_dir(base_dir)
    fname = f"{base_dir}/distances.pickle"
    if os.path.exists(fname):
        print(f"{fname} exists")  # debug
        neigh_dist_and_indices = pickle.load(open(fname, "rb"))
    else:
        print(f"{fname} doesn't exist")  # debug
        neigh_dist_and_indices = knn.kneighbors(X_test)
        pickle.dump(neigh_dist_and_indices, open(fname, "wb"))
    return neigh_dist_and_indices