App下載

Keras載入mnist數(shù)據(jù)集出錯怎么辦?

來源: 潮起潮落 2021-07-26 14:05:19 瀏覽數(shù) (2977)
反饋

mnist是一個簡單的計算機視覺數(shù)據(jù)集,它包含了各種手寫數(shù)字圖片。是很多學習機器學習的初學者第一個接觸到的數(shù)據(jù)集。但是有一部分的小伙伴反應Keras在mnist數(shù)據(jù)集載入的時候會出現(xiàn)報錯的問題,這里小編就這一問題進行一個解決方案的介紹:

1.找到本地keras目錄下的mnist.py文件,目錄:

F:python_enter_anaconda510Libsite-packages ensorflowpythonkerasdatasets

2.下載mnist.npz文件到本地,下載地址:

https://s3.amazonaws.com/img-datasets/mnist.npz

3.修改mnist.py文件為以下內容,并保存

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
 
from ..utils.data_utils import get_file
import numpy as np
 
def load_data(path='mnist.npz'):
    """Loads the MNIST dataset.
    # Arguments
        path: path where to cache the dataset locally
            (relative to ~/.keras/datasets).
    # Returns
        Tuple of Numpy arrays: `(x_train, y_train), (x_test, y_test)`.
    """
    path = 'E:/Data/Mnist/mnist.npz' #此處的path為你剛剛防止mnist.py的目錄。注意斜杠
    f = np.load(path)
    x_train, y_train = f['x_train'], f['y_train']
    x_test, y_test = f['x_test'], f['y_test']
    f.close()
    return (x_train, y_train), (x_test, y_test)

補充:Keras MNIST 手寫數(shù)字識別數(shù)據(jù)集

下載 MNIST 數(shù)據(jù)

1 導入相關的模塊

import keras
import numpy as np
from keras.utils import np_utils   
import os
from keras.datasets import mnist

2 第一次進行Mnist 數(shù)據(jù)的下載

 
(X_train_image ,y_train_image),(X_test_image,y_test_image) = mnist.load_data()

第一次執(zhí)行 mnist.load_data() 方法 ,程序會檢查用戶目錄下是否已經存在 MNIST 數(shù)據(jù)集文件 ,如果沒有,就會自動下載 . (所以第一次運行比較慢) .

3 查看已經下載的MNIST 數(shù)據(jù)文件

查看下載的mnist文件

4 查看MNIST數(shù)據(jù)

print('train data = ' ,len(X_train_image)) # 
print('test data = ',len(X_test_image))

查看訓練數(shù)據(jù)

1 訓練集是由 images 和 label 組成的 , images 是數(shù)字的單色數(shù)字圖像 28 x 28 的 , label 是images 對應的數(shù)字的十進制表示 .

2 顯示數(shù)字的圖像

import matplotlib.pyplot as plt
def plot_image(image):
    fig = plt.gcf() 
    fig.set_size_inches(2,2)  # 設置圖形的大小
    plt.imshow(image,cmap='binary') # 傳入圖像image ,cmap 參數(shù)設置為 binary ,以黑白灰度顯示 
    plt.show()

3 查看訓練數(shù)據(jù)中的第一個數(shù)據(jù)

plot_image(x_train_image[0])

訓練數(shù)據(jù)

查看對應的標記(真實值)

print(y_train_image[0])

運行結果 : 5

查看多項訓練數(shù)據(jù) images 與 label

上面我們只顯示了一組數(shù)據(jù)的圖像 , 下面將顯示多組手寫數(shù)字的圖像展示 ,以便我們查看數(shù)據(jù) .

def plot_images_labels_prediction(images, labels,
                                  prediction, idx, num=10):
    fig = plt.gcf()
    fig.set_size_inches(12, 14) # 設置大小
    if num > 25: num = 25
    for i in range(0, num):
        ax = plt.subplot(5, 5, 1 + i)# 分成 5 X 5 個子圖顯示, 第三個參數(shù)表示第幾個子圖
        ax.imshow(images[idx], cmap='binary')
        title = "label=" + str(labels[idx])
        if len(prediction) > 0: # 如果有預測值
            title += ",predict=" + str(prediction[idx])
 
        ax.set_title(title, fontsize=10)
        ax.set_xticks([])
        ax.set_yticks([])
        idx += 1
    plt.show()
plot_images_labels_prediction(x_train_image,y_train_image,[],0,10)

查看多項訓練數(shù)據(jù)

查看測試集 的手寫數(shù)字前十個

plot_images_labels_prediction(x_test_image,y_test_image,[],0,10)
 

查看測試集的前十個

多層感知器模型數(shù)據(jù)預處理

feature (數(shù)字圖像的特征值) 數(shù)據(jù)預處理可分為兩個步驟:

(1) 將原本的 288 X28 的數(shù)字圖像以 reshape 轉換為 一維的向量 ,其長度為 784 ,并且轉換為 float

(2) 數(shù)字圖像 image 的數(shù)字標準化

1 查看image 的shape

print("x_train_image : " ,len(x_train_image) , x_train_image.shape )
print("y_train_label : ", len(y_train_label) , y_train_label.shape)
#output : 
 
x_train_image :  60000 (60000, 28, 28)
y_train_label :  60000 (60000,)

2 將 lmage 以 reshape 轉換

# 將 image 以 reshape 轉化
 
x_Train = x_train_image.reshape(60000,784).astype('float32')
x_Test = x_test_image.reshape(10000,784).astype('float32')
 
print('x_Train : ' ,x_Train.shape)
print('x_Test' ,x_Test.shape)

3 標準化

images 的數(shù)字標準化可以提高后續(xù)訓練模型的準確率 ,因為 images 的數(shù)字 是從 0 到255 的值 ,代表圖形每一個點灰度的深淺 .

# 標準化 
x_Test_normalize = x_Test/255 
x_Train_normalize = x_Train/255

4 查看標準化后的測試集和訓練集 image

print(x_Train_normalize[0]) # 訓練集中的第一個數(shù)字的標準化
x_train_image :  60000 (60000, 28, 28)
y_train_label :  60000 (60000,)
[0.         0.         0.         0.         0.         0.
 
........................................................
 0.         0.         0.         0.         0.         0.
 0.
 0.21568628 0.6745098  0.8862745  0.99215686 0.99215686 0.99215686
 0.99215686 0.95686275 0.52156866 0.04313726 0.         0.
 0.         0.         0.         0.         0.         0.
 0.         0.         0.         0.         0.         0.
 0.         0.         0.         0.         0.53333336 0.99215686
 0.99215686 0.99215686 0.83137256 0.5294118  0.5176471  0.0627451
 
 0.         0.         0.         0.        ]

Label 數(shù)據(jù)的預處理

label 標簽字段原本是 0 ~ 9 的數(shù)字 ,必須以 One -hot Encoding 獨熱編碼 轉換為 10個 0,1 組合 ,比如 7 經過 One -hot encoding

轉換為 0000000100 ,正好就對應了輸出層的 10 個 神經元 .

# 將訓練集和測試集標簽都進行獨熱碼轉化
y_TrainOneHot = np_utils.to_categorical(y_train_label)
y_TestOneHot = np_utils.to_categorical(y_test_label)
print(y_TrainOneHot[:5]) # 查看前5項的標簽
[[0. 0. 0. 0. 0. 1. 0. 0. 0. 0.]     5
 [1. 0. 0. 0. 0. 0. 0. 0. 0. 0.]     0
 [0. 0. 0. 0. 1. 0. 0. 0. 0. 0.]     4
 [0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]     1
 [0. 0. 0. 0. 0. 0. 0. 0. 0. 1.]]    9

Keras 多元感知器識別 MNIST 手寫數(shù)字圖像的介紹

1 我們將將建立如圖所示的多層感知器模型

多層感知器模型

2 建立model 后 ,必須先訓練model 才能進行預測(識別)這些手寫數(shù)字 .

模型的訓練和預測

數(shù)據(jù)的預處理我們已經處理完了. 包含 數(shù)據(jù)集 輸入(數(shù)字圖像)的標準化 , label的one-hot encoding

下面我們將建立模型

我們將建立多層感知器模型 ,輸入層 共有784 個神經元 ,hodden layer 有 256 個neure ,輸出層用 10 個神經元 .

1 導入相關模塊

from keras.models import Sequential
from keras.layers import Dense

2 建立 Sequence 模型

 
# 建立Sequential 模型
model = Sequential()

3 建立 "輸入層" 和 "隱藏層"

使用 model,add() 方法加入 Dense 神經網(wǎng)絡層 .

model.add(Dense(units=256,
          input_dim =784,
          keras_initializer='normal',
          activation='relu')
          )
參數(shù) 說明
units =256 定義"隱藏層"神經元的個數(shù)為256
input_dim 設置輸入層神經元個數(shù)為 784
kernel_initialize='normal' 使用正態(tài)分布的隨機數(shù)初始化weight和bias
activation 激勵函數(shù)為 relu

4 建立輸出層

model.add(Dense(
    units=10,
    kernel_initializer='normal',
    activation='softmax'
))
 
參數(shù) 說明
units 定義"輸出層"神經元個數(shù)為10
kernel_initializer='normal' 同上
activation='softmax 激活函數(shù) softmax

5 查看模型的摘要

print(model.summary())

param 的計算是 上一次的神經元個數(shù) * 本層神經元個數(shù) + 本層神經元個數(shù) .

進行訓練

1 定義訓練方式

model.compile(loss='categorical_crossentropy' ,optimizer='adam',metrics=['accuracy'])

loss (損失函數(shù)) : 設置損失函數(shù), 這里使用的是交叉熵 .

optimizer : 優(yōu)化器的選擇,可以讓訓練更快的收斂

metrics : 設置評估模型的方式是準確率

開始訓練 2

train_history = model.fit(x=x_Train_normalize,y=y_TrainOneHot,validation_split=0.2 ,
                          epoch=10,batch_size=200,verbose=2)
 

使用 model.fit() 進行訓練 , 訓練過程會存儲在 train_history 變量中 .

(1)輸入訓練數(shù)據(jù)參數(shù)

x = x_Train_normalize

y = y_TrainOneHot

(2)設置訓練集和驗證集的數(shù)據(jù)比例

validation_split=0.2 8 :2 = 訓練集 : 驗證集

(3) 設置訓練周期 和 每一批次項數(shù)

epoch=10,batch_size=200

(4) 顯示訓練過程

verbose = 2

3 建立show_train_history 顯示訓練過程

def show_train_history(train_history,train,validation) :
 
    plt.plot(train_history.history[train])
    plt.plot(train_history.history[validation])
    plt.title("Train_history")
    plt.ylabel(train)
    plt.xlabel('Epoch')
    plt.legend(['train','validation'],loc='upper left')
    plt.show()

模型準確率


測試數(shù)據(jù)評估模型準確率

scores = model.evaluate(x_Test_normalize,y_TestOneHot)
print()
print('accuracy=',scores[1] )

accuracy= 0.9769

進行預測

通過之前的步驟, 我們建立了模型, 并且完成了模型訓練 ,準確率達到可以接受的 0.97 . 接下來我們將使用此模型進行預測.

1 執(zhí)行預測

prediction = model.predict_classes(x_Test)
print(prediction)

result : [7 2 1 ... 4 5 6]

2 顯示 10 項預測結果

plot_images_labels_prediction(x_test_image,y_test_label,prediction,idx=340)

顯示預測結果

我們可以看到 第一個數(shù)字 label 是 5 結果預測成 3 了.

顯示混淆矩陣

上面我們在預測到第340 個測試集中的數(shù)字5 時 ,卻被錯誤的預測成了 3 .如果想要更進一步的知道我們所建立的模型中哪些 數(shù)字的預測準確率更高 , 哪些數(shù)字會容忍混淆 .

混淆矩陣 也稱為 誤差矩陣.

1 使用Pandas 建立混淆矩陣 .

showMetrix = pd.crosstab(y_test_label,prediction,colnames=['label',],rownames=['predict'])
print(showMetrix)
label      0     1     2    3    4    5    6    7    8    9
predict                                                    
0        971     0     1    1    1    0    2    1    3    0
1          0  1124     4    0    0    1    2    0    4    0
2          5     0  1009    2    1    0    3    4    8    0
3          0     0     5  993    0    1    0    3    4    4
4          1     0     5    1  961    0    3    0    3    8
5          3     0     0   16    1  852    7    2    8    3
6          5     3     3    1    3    3  939    0    1    0
7          0     5    13    7    1    0    0  988    5    9
8          4     0     3    7    1    1    1    2  954    1
9          3     6     0   11    7    2    1    4    4  971

2 使用DataFrame

df = pd.DataFrame({'label ':y_test_label, 'predict':prediction})
print(df)
      label   predict
0          7        7
1          2        2
2          1        1
3          0        0
4          4        4
5          1        1
6          4        4
7          9        9
8          5        5
9          9        9
10         0        0
11         6        6
12         9        9
13         0        0
14         1        1
15         5        5
16         9        9
17         7        7
18         3        3
19         4        4
20         9        9
21         6        6
22         6        6
23         5        5
24         4        4
25         0        0
26         7        7
27         4        4
28         0        0
29         1        1
...      ...      ...
9970       5        5
9971       2        2
9972       4        4
9973       9        9
9974       4        4
9975       3        3
9976       6        6
9977       4        4
9978       1        1
9979       7        7
9980       2        2
9981       6        6
9982       5        6
9983       0        0
9984       1        1
9985       2        2
9986       3        3
9987       4        4
9988       5        5
9989       6        6
9990       7        7
9991       8        8
9992       9        9
9993       0        0
9994       1        1
9995       2        2
9996       3        3
9997       4        4
9998       5        5
9999       6        6

隱藏層增加為 1000個神經元

model.add(Dense(units=1000,
                input_dim=784,
                kernel_initializer='normal',
                activation='relu'))

hidden layer 神經元的增大,參數(shù)也增多了, 所以訓練model的時間也變慢了.

加入 Dropout 功能避免過度擬合

 
# 建立Sequential 模型
model = Sequential()
 
model.add(Dense(units=1000,
                input_dim=784,
                kernel_initializer='normal',
                activation='relu'))
model.add(Dropout(0.5)) # 加入Dropout 
model.add(Dense(units=10,
                kernel_initializer='normal',
                activation='softmax'))

驗證準確率

訓練的準確率 和 驗證的準確率 差距變小了 .

建立多層感知器模型包含兩層隱藏層

 
# 建立Sequential 模型
model = Sequential()
# 輸入層 +" 隱藏層"1 
model.add(Dense(units=1000,
                input_dim=784,
                kernel_initializer='normal',
                activation='relu'))
model.add(Dropout(0.5)) # 加入Dropout
# " 隱藏層"2
model.add(Dense(units=1000,
                kernel_initializer='normal',
                activation='relu'))
model.add(Dropout(0.5)) # 加入Dropout
# " 輸出層" 
model.add(Dense(units=10,
                kernel_initializer='normal',
                activation='softmax'))
 
print(model.summary())

運行結果

代碼:

import tensorflow as tf
import keras
import matplotlib.pyplot as plt
import numpy as np
from keras.utils import np_utils
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
import pandas as pd
import os
 
np.random.seed(10)
os.environ["CUDA_VISIBLE_DEVICES"] = "2"
 
(x_train_image ,y_train_label),(x_test_image,y_test_label) = mnist.load_data()
 
#
# print('train data = ' ,len(X_train_image)) #
# print('test data = ',len(X_test_image))
 
def plot_image(image):
    fig = plt.gcf()
    fig.set_size_inches(2,2)  # 設置圖形的大小
    plt.imshow(image,cmap='binary') # 傳入圖像image ,cmap 參數(shù)設置為 binary ,以黑白灰度顯示
    plt.show()
def plot_images_labels_prediction(images, labels,
                                  prediction, idx, num=10):
    fig = plt.gcf()
    fig.set_size_inches(12, 14)
    if num > 25: num = 25
    for i in range(0, num):
        ax = plt.subplot(5, 5, 1 + i)# 分成 5 X 5 個子圖顯示, 第三個參數(shù)表示第幾個子圖
        ax.imshow(images[idx], cmap='binary')
        title = "label=" + str(labels[idx])
        if len(prediction) > 0:
            title += ",predict=" + str(prediction[idx])
 
        ax.set_title(title, fontsize=10)
        ax.set_xticks([])
        ax.set_yticks([])
        idx += 1
    plt.show()
 
def show_train_history(train_history,train,validation) :
 
    plt.plot(train_history.history[train])
    plt.plot(train_history.history[validation])
    plt.title("Train_history")
    plt.ylabel(train)
    plt.xlabel('Epoch')
    plt.legend(['train','validation'],loc='upper left')
    plt.show()
 
# plot_images_labels_prediction(x_train_image,y_train_image,[],0,10)
#
# plot_images_labels_prediction(x_test_image,y_test_image,[],0,10)
print("x_train_image : " ,len(x_train_image) , x_train_image.shape )
print("y_train_label : ", len(y_train_label) , y_train_label.shape)
# 將 image 以 reshape 轉化
 
x_Train = x_train_image.reshape(60000,784).astype('float32')
x_Test = x_test_image.reshape(10000,784).astype('float32')
 
# print('x_Train : ' ,x_Train.shape)
# print('x_Test' ,x_Test.shape)
# 標準化
x_Test_normalize = x_Test/255
x_Train_normalize = x_Train/255
 
# print(x_Train_normalize[0]) # 訓練集中的第一個數(shù)字的標準化
# 將訓練集和測試集標簽都進行獨熱碼轉化
y_TrainOneHot = np_utils.to_categorical(y_train_label)
y_TestOneHot = np_utils.to_categorical(y_test_label)
print(y_TrainOneHot[:5]) # 查看前5項的標簽
 
# 建立Sequential 模型
model = Sequential()
model.add(Dense(units=1000,
                input_dim=784,
                kernel_initializer='normal',
                activation='relu'))
model.add(Dropout(0.5)) # 加入Dropout
# " 隱藏層"2
model.add(Dense(units=1000,
                kernel_initializer='normal',
                activation='relu'))
model.add(Dropout(0.5)) # 加入Dropout
 
model.add(Dense(units=10,
                kernel_initializer='normal',
                activation='softmax'))
print(model.summary())
 
# 訓練方式
model.compile(loss='categorical_crossentropy' ,optimizer='adam',metrics=['accuracy'])
# 開始訓練
train_history =model.fit(x=x_Train_normalize,
                         y=y_TrainOneHot,validation_split=0.2,
                         epochs=10, batch_size=200,verbose=2)
 
show_train_history(train_history,'acc','val_acc')
scores = model.evaluate(x_Test_normalize,y_TestOneHot)
print()
print('accuracy=',scores[1] )
prediction = model.predict_classes(x_Test)
print(prediction)
plot_images_labels_prediction(x_test_image,y_test_label,prediction,idx=340)
showMetrix = pd.crosstab(y_test_label,prediction,colnames=['label',],rownames=['predict'])
print(showMetrix)
df = pd.DataFrame({'label ':y_test_label, 'predict':prediction})
print(df)
 
#
#
# plot_image(x_train_image[0])
#
# print(y_train_image[0])

代碼2:

import numpy as np
from keras.models import Sequential
from keras.layers import Dense , Dropout ,Deconv2D
from keras.utils import np_utils
from keras.datasets import mnist
from keras.optimizers import SGD
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "2"
def load_data():
    (x_train,y_train),(x_test,y_test) = mnist.load_data()
    number = 10000
    x_train = x_train[0:number]
    y_train = y_train[0:number]
 
    x_train =x_train.reshape(number,28*28)
    x_test = x_test.reshape(x_test.shape[0],28*28)
    x_train = x_train.astype('float32')
    x_test = x_test.astype('float32')
    y_train = np_utils.to_categorical(y_train,10)
    y_test = np_utils.to_categorical(y_test,10)
    x_train = x_train/255
    x_test = x_test /255
    return (x_train,y_train),(x_test,y_test)
(x_train,y_train),(x_test,y_test) = load_data()
 
model = Sequential()
model.add(Dense(input_dim=28*28,units=689,activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(units=689,activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(units=689,activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(output_dim=10,activation='softmax'))
model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy'])
model.fit(x_train,y_train,batch_size=10000,epochs=20)
res1 = model.evaluate(x_train,y_train,batch_size=10000)
print("
 Train Acc :",res1[1])
res2 = model.evaluate(x_test,y_test,batch_size=10000)
print("
 Test Acc :",res2[1])

以上就是Keras載入mnist數(shù)據(jù)集的的文章的全部內容了,希望能給大家一個參考,也希望大家多多支持W3Cschool。



0 人點贊