keras实战-入门之手写数字识别DNN

手写数字识别

属于hello world级别的练习，就不多说了

import numpy as np
np.random.seed(11)

from keras.datasets import mnist
from keras.utils import np_utils

Using TensorFlow backend.

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

#加载数据集
(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))

train data= 60000
test data= 10000

print('y_train_label',x_train_image.shape)
print('y_train_label',y_train_label.shape)

y_train_label (60000, 28, 28)
y_train_label (60000,)

#看看字体图片
import matplotlib.pyplot as plt
def show_image(image):
    fig=plt.figure(figsize=(4,4))
    plt.imshow(image,cmap='binary')
    plt.show()

show_image(x_train_image[0])

y_train_label[0]

5

#处理数据，把他展平了
x_train=x_train_image.reshape(60000,-1).astype('float32')
x_test=x_test_image.reshape(10000,-1).astype('float32')
print(x_train.shape)
print(x_test.shape)

(60000, 784)
(10000, 784)

#归一化数据
x_train_normalize=x_train/255
x_test_normalize=x_test/255

#one hot编码
y_train_oneHot=np_utils.to_categorical(y_train_label)
y_test_oneHot=np_utils.to_categorical(y_test_label)

y_train_oneHot[:3]

array([[0., 0., 0., 0., 0., 1., 0., 0., 0., 0.],
       [1., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
       [0., 0., 0., 0., 1., 0., 0., 0., 0., 0.]])

#用了最简单的一层，当然也可以有其他优化
model=Sequential()
model.add(Dense(units=256,input_dim=784,kernel_initializer='normal',activation='relu'))
model.add(Dense(units=10,kernel_initializer='normal',activation='softmax'))

WARNING:tensorflow:From F:\Anaconda3\lib\site-packages\tensorflow\python\framework\op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.
Instructions for updating:
Colocations handled automatically by placer.

#看看结构
print(model.summary())

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_1 (Dense)              (None, 256)               200960    
_________________________________________________________________
dense_2 (Dense)              (None, 10)                2570      
=================================================================
Total params: 203,530
Trainable params: 203,530
Non-trainable params: 0
_________________________________________________________________
None

#编译模型，交叉熵，优化器SGD，要看准确率
model.compile(loss='categorical_crossentropy',optimizer='sgd',metrics=['accuracy'])

train_history=model.fit(x_train_normalize,y_train_oneHot,validation_split=0.2,epochs=10,batch_size=256,verbose=2)

WARNING:tensorflow:From F:\Anaconda3\lib\site-packages\tensorflow\python\ops\math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.cast instead.
Train on 48000 samples, validate on 12000 samples
Epoch 1/10
 - 3s - loss: 1.9122 - acc: 0.5346 - val_loss: 1.4928 - val_acc: 0.7507
Epoch 2/10
 - 1s - loss: 1.1975 - acc: 0.7794 - val_loss: 0.9231 - val_acc: 0.8319
Epoch 3/10
 - 1s - loss: 0.8179 - acc: 0.8295 - val_loss: 0.6780 - val_acc: 0.8607
Epoch 4/10
 - 1s - loss: 0.6491 - acc: 0.8529 - val_loss: 0.5606 - val_acc: 0.8740
Epoch 5/10
 - 1s - loss: 0.5596 - acc: 0.8658 - val_loss: 0.4939 - val_acc: 0.8831
Epoch 6/10
 - 1s - loss: 0.5045 - acc: 0.8728 - val_loss: 0.4505 - val_acc: 0.8888
Epoch 7/10
 - 1s - loss: 0.4668 - acc: 0.8796 - val_loss: 0.4207 - val_acc: 0.8937
Epoch 8/10
 - 1s - loss: 0.4395 - acc: 0.8845 - val_loss: 0.3986 - val_acc: 0.8976
Epoch 9/10
 - 1s - loss: 0.4184 - acc: 0.8885 - val_loss: 0.3815 - val_acc: 0.8993
Epoch 10/10
 - 1s - loss: 0.4016 - acc: 0.8919 - val_loss: 0.3680 - val_acc: 0.9029

loss, accuracy = model.evaluate(x_test_normalize, y_test_oneHot)
print('test loss: ', loss)
print('test accuracy: ', accuracy)

10000/10000 [==============================] - 1s 63us/step
test loss:  0.3700620859503746
test accuracy:  0.9006

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

#查看准确率曲线，是否过拟合
show_train_history(train_history,'acc','val_acc')

#loss也可以看到
show_train_history(train_history,'loss','val_loss')

def show_images_labels_prediction(images,labels,prediction,idx,num=10):
    flig=plt.figure(figsize=(12,14))
    if num>25:
        num=25
    for i in range(0,num):
        ax=plt.subplot(5,5,1+i)
        ax.imshow(images[idx],cmap='binary')
        title='labels='+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()

#看下具体训练集和标签
show_images_labels_prediction(x_train_image,y_train_label,[],0,10)

#进行类别预测
prediction=model.predict_classes(x_test_normalize)

prediction

array([7, 2, 1, ..., 4, 5, 6], dtype=int64)

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

#可以查看预测和标签混淆矩阵，来观察哪些预测有问题，对角线是预测对的，其他就是预测错的
#混淆矩阵
import pandas as pd
pd.crosstab(y_test_label,prediction,rownames=['label'],colnames=['predict'])

predict	0	1	2	3	4	5	6	7	8	9
label
0	952	0	4	2	0	4	12	1	5	0
1	0	1108	2	3	1	2	4	1	14	0
2	12	7	892	17	17	2	21	21	37	6
3	6	1	18	893	1	41	3	16	22	9
4	2	3	5	0	901	1	15	2	11	42
5	15	4	5	48	9	739	17	8	41	6
6	17	3	9	2	15	18	889	1	4	0
7	4	20	31	2	12	0	0	919	5	35
8	8	9	12	34	10	25	14	12	836	14
9	13	9	7	11	50	14	0	20	8	877

#上图可以看到5被识别为3的有48个，那我们把他们找几个出来看看
df=pd.DataFrame({'label':y_test_label,'predict':prediction})
data=df[(df.label==5)&(df.predict==3)]

data[0:5]

	label	predict
261	5	3
340	5	3
502	5	3
857	5	3
1082	5	3

#显示几个看看
for i in data[0:5].index:
    show_images_labels_prediction(x_test_image,y_test_label,prediction,idx=i,num=1)

#上面貌似5的准确率还是有待提高，再来看看2预测成1的吧
data=df[(df.label==2)&(df.predict==1)]

data

	label	predict
536	2	1
1409	2	1
2269	2	1
2433	2	1
3946	2	1
4812	2	1
9168	2	1

#显示几个看看
for i in data[0:5].index:
    show_images_labels_prediction(x_test_image,y_test_label,prediction,idx=i,num=1)

可见全连接神经网络识别还是有问题，后面用CNN应该会比较好

好了，今天就到这里了，希望对学习理解有帮助，大神看见勿喷，仅为自己的学习理解，能力有限，请多包涵，侵删。