F:\python_enter_anaconda510\Lib\site-packages\tensorflow\python\keras\datasets
https://s3.amazonaws.com/img-datasets/mnist.npz
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(bǔ)充：Keras MNIST 手寫(xiě)數(shù)字識(shí)別數(shù)據(jù)集
1 導(dǎo)入相關(guān)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(jìn)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(huì)檢查用戶目錄下是否已經(jīng)存在 MNIST 數(shù)據(jù)集文件 ,如果沒(méi)有,就會(huì)自動(dòng)下載 . (所以第一次運(yùn)行比較慢) .
3 查看已經(jīng)下載的MNIST 數(shù)據(jù)文件
4 查看MNIST數(shù)據(jù)
print('train data = ' ,len(X_train_image)) #
print('test data = ',len(X_test_image))
1 訓(xùn)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(duì)應(yīng)的數(shù)字的十進(jìn)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(shè)置圖形的大小
plt.imshow(image,cmap='binary') # 傳入圖像image ,cmap 參數(shù)設(shè)置為 binary ,以黑白灰度顯示
plt.show()
3 查看訓(xùn)練數(shù)據(jù)中的第一個(gè)數(shù)據(jù)
plot_image(x_train_image[0])
查看對(duì)應(yīng)的標(biāo)記(真實(shí)值)
print(y_train_image[0])
運(yùn)行結(jié)果 : 5
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(xiě)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(shè)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(gè)子圖顯示, 第三個(gè)參數(shù)表示第幾個(gè)子圖
ax.imshow(images[idx], cmap='binary')
title = "label=" + str(labels[idx])
if len(prediction) > 0: # 如果有預(yù)測(cè)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)
查看測(cè)試集 的手寫(xiě)數(shù)字前十個(gè)
plot_images_labels_prediction(x_test_image,y_test_image,[],0,10)
feature (數(shù)字圖像的特征值) 數(shù)據(jù)預(yù)處理可分為兩個(gè)步驟:
(1) 將原本的 288 X28 的數(shù)字圖像以 reshape 轉(zhuǎn)換為 一維的向量 ,其長(zhǎng)度為 784 ,并且轉(zhuǎn)換為 float
(2) 數(shù)字圖像 image 的數(shù)字標(biāo)準(zhǔn)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(zhuǎn)換
# 將 image 以 reshape 轉(zhuǎn)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(biāo)準(zhǔn)化
images 的數(shù)字標(biāo)準(zhǔn)化可以提高后續(xù)訓(xùn)練模型的準(zhǔn)確率 ,因?yàn)?images 的數(shù)字 是從 0 到255 的值 ,代表圖形每一個(gè)點(diǎn)灰度的深淺 .
# 標(biāo)準(zhǔn)化 x_Test_normalize = x_Test/255 x_Train_normalize = x_Train/255
4 查看標(biāo)準(zhǔn)化后的測(cè)試集和訓(xùn)練集 image
print(x_Train_normalize[0]) # 訓(xùn)練集中的第一個(gè)數(shù)字的標(biāo)準(zhǔn)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 標(biāo)簽字段原本是 0 ~ 9 的數(shù)字 ,必須以 One -hot Encoding 獨(dú)熱編碼 轉(zhuǎn)換為 10個(gè) 0,1 組合 ,比如 7 經(jīng)過(guò) One -hot encoding
轉(zhuǎn)換為 0000000100 ,正好就對(duì)應(yīng)了輸出層的 10 個(gè) 神經(jīng)元 .
# 將訓(xùn)練集和測(cè)試集標(biāo)簽都進(jìn)行獨(dú)熱碼轉(zhuǎn)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(xiàng)的標(biāo)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
1 我們將將建立如圖所示的多層感知器模型
2 建立model 后 ,必須先訓(xùn)練model 才能進(jìn)行預(yù)測(cè)(識(shí)別)這些手寫(xiě)數(shù)字 .
數(shù)據(jù)的預(yù)處理我們已經(jīng)處理完了. 包含 數(shù)據(jù)集 輸入(數(shù)字圖像)的標(biāo)準(zhǔn)化 , label的one-hot encoding
我們將建立多層感知器模型 ,輸入層 共有784 個(gè)神經(jīng)元 ,hodden layer 有 256 個(gè)neure ,輸出層用 10 個(gè)神經(jīng)元 .
1 導(dǎo)入相關(guān)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(jīng)網(wǎng)絡(luò)層 .
model.add(Dense(units=256,
input_dim =784,
keras_initializer='normal',
activation='relu')
)
| 參數(shù) | 說(shuō)明 |
| units =256 | 定義"隱藏層"神經(jīng)元的個(gè)數(shù)為256 |
| input_dim | 設(shè)置輸入層神經(jīng)元個(gè)數(shù)為 784 |
| kernel_initialize='normal' | 使用正態(tài)分布的隨機(jī)數(shù)初始化weight和bias |
| activation | 激勵(lì)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(shuō)明 |
| units | 定義"輸出層"神經(jīng)元個(gè)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(jì)算是 上一次的神經(jīng)元個(gè)數(shù) * 本層神經(jīng)元個(gè)數(shù) + 本層神經(jīng)元個(gè)數(shù) .
1 定義訓(xùn)練方式
model.compile(loss='categorical_crossentropy' ,optimizer='adam',metrics=['accuracy'])
loss (損失函數(shù)) : 設(shè)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(xùn)練更快的收斂
metrics : 設(shè)置評(píng)估模型的方式是準(zhǔn)確率
開(kāi)始訓(xùn)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(jìn)行訓(xùn)練 , 訓(xùn)練過(guò)程會(huì)存儲(chǔ)在 train_history 變量中 .
(1)輸入訓(xùn)練數(shù)據(jù)參數(shù)
x = x_Train_normalize
y = y_TrainOneHot
(2)設(shè)置訓(xùn)練集和驗(yàn)證集的數(shù)據(jù)比例
validation_split=0.2 8 :2 = 訓(xùn)練集 : 驗(yàn)證集
(3) 設(shè)置訓(xùn)練周期 和 每一批次項(xiàng)數(shù)
epoch=10,batch_size=200
(4) 顯示訓(xùn)練過(guò)程
verbose = 2
3 建立show_train_history 顯示訓(xùn)練過(guò)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()
測(cè)試數(shù)據(jù)評(píng)估模型準(zhǔn)確率
scores = model.evaluate(x_Test_normalize,y_TestOneHot)
print()
print('accuracy=',scores[1] )
accuracy= 0.9769
通過(guò)之前的步驟, 我們建立了模型, 并且完成了模型訓(xùn)練 ,準(zhǔn)確率達(dá)到可以接受的 0.97 . 接下來(lái)我們將使用此模型進(jìn)行預(yù)測(cè).
1 執(zhí)行預(yù)測(cè)
prediction = model.predict_classes(x_Test) print(prediction)
result : [7 2 1 ... 4 5 6]
2 顯示 10 項(xiàng)預(yù)測(cè)結(jié)果
plot_images_labels_prediction(x_test_image,y_test_label,prediction,idx=340)
我們可以看到 第一個(gè)數(shù)字 label 是 5 結(jié)果預(yù)測(cè)成 3 了.
上面我們?cè)陬A(yù)測(cè)到第340 個(gè)測(cè)試集中的數(shù)字5 時(shí) ,卻被錯(cuò)誤的預(yù)測(cè)成了 3 .如果想要更進(jìn)一步的知道我們所建立的模型中哪些 數(shù)字的預(yù)測(cè)準(zhǔn)確率更高 , 哪些數(shù)字會(huì)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
model.add(Dense(units=1000,
input_dim=784,
kernel_initializer='normal',
activation='relu'))
hidden layer 神經(jīng)元的增大,參數(shù)也增多了, 所以訓(xùn)練model的時(shí)間也變慢了.
加入 Dropout 功能避免過(guò)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'))
訓(xùn)練的準(zhǔn)確率 和 驗(yàn)證的準(zhǔn)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())
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(shè)置圖形的大小
plt.imshow(image,cmap='binary') # 傳入圖像image ,cmap 參數(shù)設(shè)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(gè)子圖顯示, 第三個(gè)參數(shù)表示第幾個(gè)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(zhuǎn)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(biāo)準(zhǔn)化
x_Test_normalize = x_Test/255
x_Train_normalize = x_Train/255
# print(x_Train_normalize[0]) # 訓(xùn)練集中的第一個(gè)數(shù)字的標(biāo)準(zhǔn)化
# 將訓(xùn)練集和測(cè)試集標(biāo)簽都進(jìn)行獨(dú)熱碼轉(zhuǎn)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(xiàng)的標(biāo)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(xùn)練方式
model.compile(loss='categorical_crossentropy' ,optimizer='adam',metrics=['accuracy'])
# 開(kāi)始訓(xùn)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("\n Train Acc :",res1[1])
res2 = model.evaluate(x_test,y_test,batch_size=10000)
print("\n Test Acc :",res2[1])
以上為個(gè)人經(jīng)驗(yàn)，希望能給大家一個(gè)參考，也希望大家多多支持腳本之家。
標(biāo)簽：常德 黑龍江 潛江 阿里 株洲 通遼 呂梁 銅川
巨人網(wǎng)絡(luò)通訊聲明：本文標(biāo)題《如何解決Keras載入mnist數(shù)據(jù)集出錯(cuò)的問(wèn)題》，本文關(guān)鍵詞 如何,解決,Keras,載入,mnist,；如發(fā)現(xiàn)本文內(nèi)容存在版權(quán)問(wèn)題，煩請(qǐng)?zhí)峁┫嚓P(guān)信息告之我們，我們將及時(shí)溝通與處理。本站內(nèi)容系統(tǒng)采集于網(wǎng)絡(luò)，涉及言論、版權(quán)與本站無(wú)關(guān)。
