У алгоритма KNN из проекта OCR возникли проблемы с определенными цифрами ⇐ Python

[Anonymous]

Я пытаюсь написать проект OCR для идентификации цифр без использования каких-либо дополнительных библиотек. Я использую Pygame для рисования и визуализации числа для идентификации. Мне удалось заставить все это работать, но по какой-то странной причине некоторые конкретные числа, такие как 9 или 6, никогда не распознаются алгоритмом (даже когда я устанавливаю очень высокое значение k, ни одна из возможностей не является этим числом). Я понятия не имею, почему, и буду очень признателен за любую помощь
import time
import pygame
from sys import exit
import random
import math
number_comparisons = 60000
DATA_DIR = r"C:/Users/----/Downloads/OCR/"
TEST_DIR = r"C:/Users/----/Downloads/OCR/test/"
TEST_DATA_FILENAME = DATA_DIR + "t10k-images.idx3-ubyte"
TEST_LABELS_FILENAME = DATA_DIR + "t10k-labels.idx1-ubyte"
TRAIN_DATA_FILENAME = DATA_DIR + "train-images.idx3-ubyte"
TRAIN_LABELS_FILENAME = DATA_DIR + "train-labels.idx1-ubyte"
start_time = time.time()
DEBUG = True
#starts pygame (images and sounds)
pygame.init()
pygame.mixer.init()
#create screen variable
width = 784
height = 850
screen = pygame.display.set_mode((width,height))

#ponerle titulo
pygame.display.set_caption("Optical Character Recognition")

#reloj/framerate
clock = pygame.time.Clock()

def read_labels(filename,n_max_labels = None):
labels = [] #variable que guarda todas las imagenes
with open(filename, "rb") as f: #abrir el fichero filename como f, y leerlo en binario ("rb")
_ =f.read(4) #numero inutil (representa algo que no necesitamos)
#los siguientes 12 bytes representan el numero de imagenes, el numero de filas y de columnas
n_labels = bytes_to_int(f.read(4))
if n_max_labels:
n_labels = n_max_labels
for label_idx in range(n_labels):
label = f.read(1)
labels.append(label)

return labels

count = 0

def read_images(filename,n_max_images = None):
global count
images = [] #variable que guarda todas las imagenes
with open(filename, "rb") as f: #abrir el fichero filename como f, y leerlo en binario ("rb")
_ =f.read(4) #numero inutil (representa algo que no necesitamos)
#los siguientes 12 bytes representan el numero de imagenes, el numero de filas y de columnas
n_images = bytes_to_int(f.read(4))
if n_max_images:
n_images = n_max_images
n_rows = bytes_to_int(f.read(4))
n_columns = bytes_to_int(f.read(4))
for image_idx in range(n_images):
image = []#variable que guarda la imagen actual
for row_idx in range(n_rows):
row = []#variable que guarda la columna actual
for column_idx in range(n_columns):
count += 1
pixel = f.read(1) #leemos el pixel actual de 8 bits y lo apendizamos a la row
row.append(pixel)
image.append(row)#metemos la row en la image
images.append(image)#metemos la image en el conjunto de images
return images

def bytes_to_int(byte_data):
if byte_data == 0:
return 0
elif byte_data == 200:
return 255
elif byte_data == 200:
return 200
else:
return int.from_bytes(byte_data,"big")

def pasar_lista_unidimensional(X):
lista = []
for i in range(len(X)):
for j in range(len(X[0])):
lista.append(X[j])
return [lista]
def pasar_lista_unidimensional2(X):
return [aplanar_lista(sample) for sample in X]
def aplanar_lista(l):
return [pixel for sublist in l for pixel in sublist]
def dist(x,y):
temp2 = []
for x_i,y_i in zip(x,y):
temp2.append((bytes_to_int(x_i) - bytes_to_int(y_i)) **2)
return sum(temp2)**0.5
return sum((bytes_to_int(x_i) - bytes_to_int(y_i)) **2 for x_i,y_i in zip(x,y))**0.5 #distancia euclides
def distancia_entre_samples(X_train,test_sample):
return [dist(train_sample,test_sample) for train_sample in X_train] #por todas las imagenes, calculamos su distancia arriba

def most_frequent_element(list):
return max(list, key= list.count)

def knn(X_train,y_train,X_test, k = 3):
y_pred = [] #la prediccion que tenemos a los x_test
print("Using the knn algorithm to determine k nearest numbers to drawing...")
for test_sample_idx,test_sample in enumerate(X_test):
training_distances = distancia_entre_samples(X_train,test_sample) #queremos conseguir las distancias a todos los puntos
sorted_distance_indices = [
pair[0]
for pair in sorted(enumerate(training_distances), key = lambda x: x[1]) ]#escogemos la menor distancia
candidates = [bytes_to_int(y_train[idx]) for idx in sorted_distance_indices[:k]] # k mejores candidatos
print("Top k choices were", candidates)
top_candidate = most_frequent_element(candidates)
y_pred.append(top_candidate) #apuntamos a predicción
return y_pred
def main():
global X_test
print("Reading training files...")
#"X" es igual al dataset y "y" es el label asignado
X_train = read_images(TRAIN_DATA_FILENAME,number_comparisons)
y_train = read_labels(TRAIN_LABELS_FILENAME,number_comparisons)
y_test = read_labels(TEST_LABELS_FILENAME,1)

print("Converting drawing to 2D grid")
X_train = pasar_lista_unidimensional2(X_train) #queremos pasar la matriz de valores a una matriz unidimensional
y_pred = knn(X_train,y_train,X_test,15)
print("The number you have just written is: " ,y_pred)
print("Number of iterations: ", count)
image_array = []
font = pygame.font.Font(None, 24)
if __name__ == "__main__":
pass
for i in range(0,784,28):
pygame.draw.line(screen,"white",(0,i),(784,i))
for j in range(0,width,28):
image_array.append([0]*28)
pygame.draw.line(screen,"white",(j,0),(j,784))
print(type(image_array),type(image_array[0]),type(image_array[0][0]))
def draw(x,y):
pygame.draw.rect(screen,"white",(x-x%28,y-y%28,28,28))
image_array[math.trunc(y/28)][math.trunc(x/28)] = 200

pygame.draw.rect(screen,"white",(x-x%28+28,y-y%28,28,28))
if math.trunc(x/28)+1 < 28: image_array[math.trunc(y/28)][math.trunc(x/28)+1] = 200

pygame.draw.rect(screen,"white",(x-x%28-28,y-y%28,28,28))
if math.trunc(x/28)-1 >= 0: image_array[math.trunc(y/28)][math.trunc(x/28)-1] = 200

pygame.draw.rect(screen,"white",(x-x%28,y-y%28+28,28,28))
if math.trunc(y/28)+1 < 28: image_array[math.trunc(y/28)+1][math.trunc(x/28)] = 200

pygame.draw.rect(screen,"white",(x-x%28,y-y%28-28,28,28))
if math.trunc(y/28)-1 >= 0: image_array[math.trunc(y/28)-1][math.trunc(x/28)] = 200

button = False
text = font.render("Guess Number", True, (0, 0, 0))
while True:
#para que se pueda cerrar
for event in pygame.event.get():
if event.type == pygame.QUIT:
pygame.QUIT()
exit()
if event.type == pygame.MOUSEBUTTONDOWN:
button = True

elif event.type == pygame.MOUSEBUTTONUP:
button = False
if button == True:
pos = pygame.mouse.get_pos()
if pos[1] < 756:
draw(pos[0],pos[1])
else:
X_test = pasar_lista_unidimensional(image_array)
print(X_test)
print("Loading...")
main()
pygame.draw.rect(screen,"red",(0,784,784,66))
pygame.display.update()

#framerate
clock.tick(60)


Подробнее здесь: https://stackoverflow.com/questions/782 ... ain-digits