def random_image(self, image_path, label_path): # line:31
image_list, label_list = read_picture.read_image_data(image_path, label_path) # line:33
sample = random.sample(range(0, len(image_list)), 32) # line:34
print('随机取样完成') # line:35
color_dict = dict() # line:36
for real_img_index in sample: # line:37
color_dict[real_img_index] = random.randint(0, 2) # line:38
for img_index in range(len(sample)): # line:39
image_label = label_list[sample[img_index]] # line:40
for i in range(random.randint(0, len(image_list) // 2),
len(image_list)): # line:43
if label_list[i] == image_label and i not in color_dict: # line:44
sample.append(i) # line:45
color_dict[i] = color_dict[sample[img_index]] # line:46
break # line:47
random.shuffle(sample) # line:48
image_pair = list() # line:50
for img_index in range(len(sample)): # line:51
real_img_index = sample[img_index] # line:52
image_data = image_list[real_img_index] # line:53
image_label = label_list[real_img_index] # line:54
color = color_dict[real_img_index] # line:55
read_picture.image_save(image_data, color, "../auto_grader/image/" + str(img_index) + ".png") # line:57
image_pair.append([color, image_label]) # line:58
print('图片显示完毕') # line:59
return image_pair # line:60
def random_image(O000O0OOO0OOOO000, O0OOOO00OO0O00000,
OOO0OOO0OO00000OO): #line:31
OO00OOO00OO00O0O0, O0000O000000O000O = read_picture.read_image_data(
O0OOOO00OO0O00000, OOO0OOO0OO00000OO) #line:33
O00O000O00000O00O = random.sample(range(0, len(OO00OOO00OO00O0O0)),
32) #line:34
print('随机取样完成') #line:35
O0OOO00OO0O00O000 = dict() #line:36
for OO0OO0OOOO0OOO0OO in O00O000O00000O00O: #line:37
O0OOO00OO0O00O000[OO0OO0OOOO0OOO0OO] = random.randint(0,
2) #line:38
for O0000O0O0O0O000O0 in range(len(O00O000O00000O00O)): #line:39
OOO0O000000O000OO = O0000O000000O000O[
O00O000O00000O00O[O0000O0O0O0O000O0]] #line:40
for OOO0000OOO0OOOO00 in range(
random.randint(0,
len(OO00OOO00OO00O0O0) // 2),
len(OO00OOO00OO00O0O0)): #line:43
if O0000O000000O000O[
OOO0000OOO0OOOO00] == OOO0O000000O000OO and OOO0000OOO0OOOO00 not in O0OOO00OO0O00O000: #line:44
O00O000O00000O00O.append(OOO0000OOO0OOOO00) #line:45
O0OOO00OO0O00O000[OOO0000OOO0OOOO00] = O0OOO00OO0O00O000[
O00O000O00000O00O[O0000O0O0O0O000O0]] #line:46
break #line:47
random.shuffle(O00O000O00000O00O) #line:48
OO00O0O00O0OO0OOO = list() #line:50
for O0000O0O0O0O000O0 in range(len(O00O000O00000O00O)): #line:51
OO0OO0OOOO0OOO0OO = O00O000O00000O00O[O0000O0O0O0O000O0] #line:52
OO00OOO0OO00OOOOO = OO00OOO00OO00O0O0[OO0OO0OOOO0OOO0OO] #line:53
OOO0O000000O000OO = O0000O000000O000O[OO0OO0OOOO0OOO0OO] #line:54
O000000O00O0OO0O0 = O0OOO00OO0O00O000[OO0OO0OOOO0OOO0OO] #line:55
read_picture.image_save(OO00OOO0OO00OOOOO, O000000O00O0OO0O0,
"../auto_grader/image/" +
str(O0000O0O0O0O000O0) + ".png") #line:57
OO00O0O00O0OO0OOO.append([O000000O00O0OO0O0,
OOO0O000000O000OO]) #line:58
print('图片显示完毕') #line:59
return OO00O0O00O0OO0OOO #line:60
def identify_numbers(test_image):
train_image, train_label = read_picture.read_image_data(
'./mnist_data/train-images.idx3-ubyte',
'./mnist_data/train-labels.idx1-ubyte')
x_test, y_test = read_picture.read_image_data(
'./mnist_data/t10k-images.idx3-ubyte',
'./mnist_data/t10k-labels.idx1-ubyte')
# (train_image, train_label), (x_test, y_test) = mnist.load_data()
# train_image_arr = np.array(np.concatenate( [train_image , x_test], axis = 0 )) #array
# train_label_arr = np.array(np.concatenate( [train_label , y_test], axis = 0 ))
train_image_arr = np.array(
np.concatenate([train_image, x_test[0:5000]], axis=0)) #array
train_label_arr = np.array(
np.concatenate([train_label, y_test[0:5000]], axis=0))
# train_image_arr = np.array(train_image) # before submission, comment these two lines
# train_label_arr = np.array(train_label) # before submission, comment these two lines
large_test_image_arr = np.array(x_test)
large_test_label_arr = np.array(y_test)
test_image_arr = np.array(test_image)
# print(len(train_label_arr))
# print(len(train_image_arr))
train_image_arr = np.reshape(train_image_arr, (-1, 784))
test_image_arr = np.reshape(test_image_arr, (-1, 784))
large_test_image_arr = np.reshape(large_test_image_arr, (-1, 784))
# print(train_image_arr[0])
# train_image_arr[train_image_arr <= 0] = 0 # 转化为黑白图像
train_image_arr[train_image_arr > 0] = 255 # 转化为黑白图像
#
#
# # test_image_arr[test_image_arr <= 0] = 0 # 转化为黑白图像
test_image_arr[test_image_arr > 0] = 255 # 转化为黑白图像
#
# # large_test_image_arr[large_test_image_arr <= 0] = 0 # 转化为黑白图像
large_test_image_arr[large_test_image_arr > 0] = 255 # 转化为黑白图像
# mean_num = np.mean(train_image_arr, axis=0)
# # print(mean_num)
# train_image_arr = train_image_arr - mean_num
# large_test_image_arr = large_test_image_arr - mean_num
# test_image_arr = test_image_arr - mean_num
# pca = PCA(n_components=100)
#
# pca.fit(train_image_arr) # fit PCA with training data instead of the whole dataset
# train_image_pca = pca.transform(train_image_arr)
# test_image_pca = pca.transform(test_image_arr)
# large_test_image_pca = pca.transform(large_test_image_arr)
# train_image_pca[train_image_pca > 0] = 255 # 转化为黑白图像
# test_image_pca[test_image_pca > 0] = 255 # 转化为黑白图像
# large_test_image_pca[large_test_image_pca > 0] = 255 # 转化为黑白图像
f = kNN.KNNClassify()
f.fit(train_image_arr, train_label_arr)
# f.fit(train_image_pca, train_label_arr)
# train_image_arr_e, train_label_arr_e = f.edit(train_image_arr, train_label_arr)
# train_image_arr_e, train_label_arr_e = f.edit(train_image_pca, train_label_arr)
# train_image_arr_e = np.array(train_image_arr_e)
# train_label_arr_e = np.array(train_label_arr_e)
# edited_X = open("X_train_edited.txt", mode='wb')
# edited_y = open("y_train_edited.txt", mode='wb')
# X_dtype = train_image_arr_e.dtype
# y_dtype = train_label_arr_e.dtype
# X_shape = train_image_arr_e.shape
# y_shape = train_label_arr_e.shape
# print(X_dtype)
# print(y_dtype)
# print(X_shape)
# print(y_shape)
# edited_X.write(train_image_arr_e.tostring())
# edited_y.write(train_label_arr_e.tostring())
# train_image_arr, train_label_arr = f.edit(train_image_pca, train_label_arr)
# edited_X = open("X_train_edited.txt", mode='rb')
# edited_y = open("y_train_edited.txt", mode='rb')
# train_image_arr = np.fromstring(edited_X.read(), dtype=getattr(np, "uint8")).reshape(70000,784)
# train_label_arr = np.fromstring(edited_y.read(), dtype=getattr(np, "int32")).reshape(70000,)
y_pre_1 = f.predict_y(large_test_image_arr[5000:6000], train_image_arr,
train_label_arr)
#
print(f.score(y_pre_1, large_test_label_arr[5000:6000]))
# y_pre = f.predict_y(test_image_pca, train_image_pca, train_label_arr)
# y_pre = f.predict_y(test_image_arr, train_image_arr_e, train_label_arr_e)
y_pre = f.数字识别(test_image_arr, train_image_arr, train_label_arr)
return y_pre
if ans_max < test_ans[num_i]:
ans_max = test_ans[num_i]
ans_i = num_i
num_i += 1
ans.append(self.num_kind[ans_i])
ans = np.array(ans)
return ans
if __name__ == "__main__":
import sys
sys.path.append("../read_picture/")
import read_picture
train_image, train_label = read_picture.read_image_data(
'../../mnist_data/train-images.idx3-ubyte',
'../../mnist_data/train-labels.idx1-ubyte')
train_image_vector = np.reshape(train_image, (60000, 784))
simple_train = simple_train_one_num(train_image_vector[0:5000],
train_label[0:5000], 10, 0.1, 2.55)
simple_train.train_learn()
#构造测试集
test_image_vector = train_image_vector[5000:5100]
test_ans = train_label[5000:5100]
#计算预测
pre_ans = simple_train.predict(test_image_vector)