2-Softmax Regression

Score: 0.90971

y=softmax(xW+b)（特征缩放：特征/255）

详细分析见：TensorFlow (2): Softmax Regression识别手写数字

In [1]:
import pandas as pd

train_data=pd.read_csv('train.csv')
test_df=pd.read_csv('test.csv')
train_data.shape,test_df.shape
Out[1]:
((42000, 785), (28000, 784))
In [2]:
train_data.head()
Out[2]:
label pixel0 pixel1 pixel2 pixel3 pixel4 pixel5 pixel6 pixel7 pixel8 ... pixel774 pixel775 pixel776 pixel777 pixel778 pixel779 pixel780 pixel781 pixel782 pixel783
0 1 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
2 1 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
3 4 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0

5 rows × 785 columns

In [3]:
labels=train_data['label']
images=train_data.drop(['label'],axis=1).values
len(labels),images.shape
Out[3]:
(42000, (42000, 784))
In [4]:
labels=pd.get_dummies(labels)
print labels.shape
labels.head()

(42000, 10)
Out[4]:
0 1 2 3 4 5 6 7 8 9
0 0 1 0 0 0 0 0 0 0 0
1 1 0 0 0 0 0 0 0 0 0
2 0 1 0 0 0 0 0 0 0 0
3 0 0 0 0 1 0 0 0 0 0
4 1 0 0 0 0 0 0 0 0 0
In [5]:
labels=labels.values
labels.shape
Out[5]:
(42000, 10)
In [6]:
import numpy as np

images=np.multiply(images,1.0/255.0)
images.shape
Out[6]:
(42000, 784)
In [7]:
TRAIN_SIZE=40000
train_images=images[:TRAIN_SIZE]
train_labels=labels[:TRAIN_SIZE]
test_images=images[TRAIN_SIZE:]
test_labels=labels[TRAIN_SIZE:]

train_images.shape,len(train_labels),test_images.shape,len(test_labels)
Out[7]:
((40000, 784), 40000, (2000, 784), 2000)
In [8]:
train_images[:5]
Out[8]:
array([[ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.]])
In [9]:
train_labels[:5]
Out[9]:
array([[0, 1, 0, 0, 0, 0, 0, 0, 0, 0],
       [1, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 1, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
       [1, 0, 0, 0, 0, 0, 0, 0, 0, 0]], dtype=uint8)
In [10]:
import tensorflow as tf
x=tf.placeholder(tf.float32,[None,784])
W=tf.Variable(tf.zeros([784,10]))
b=tf.Variable(tf.zeros([10]))
y=tf.nn.softmax(tf.matmul(x,W)+b)
In [11]:
y_=tf.placeholder(tf.float32,[None,10])
cross_entropy=-tf.reduce_sum(y_*tf.log(y))
In [12]:
BATCH_SIZE=50
index_in_epoch=0
num_examples=train_images.shape[0]

def next_batch(batch_size):
    global train_images
    global train_labels
    global index_in_epoch
    
    start=index_in_epoch
    index_in_epoch+=batch_size
    
    if index_in_epoch>num_examples:
        print 'epoches copleted!'
        # 一轮完毕，shuffle数据
        perm=np.arange(num_examples) # 生成一个序列
        np.random.shuffle(perm)
        train_images=train_images[perm]
        train_labels=train_labels[perm]
        # 开始下个epoch
        start=0
        index_in_epoch=batch_size
        assert batch_size<=num_examples
    end=index_in_epoch
    return train_images[start:end],train_labels[start:end]
In [13]:
train_step=tf.train.GradientDescentOptimizer(0.001).minimize(cross_entropy)
sess=tf.InteractiveSession()
tf.global_variables_initializer().run()
In [14]:
correct_prediction=tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
accuracy=tf.reduce_mean(tf.cast(correct_prediction,'float'))
In [15]:
for i in range(3000):
    batch_xs,batch_ys=next_batch(BATCH_SIZE)
    if i%200==0:
        train_accuracy=accuracy.eval({x:batch_xs,y_:batch_ys})
        test_accuracy=accuracy.eval({x:test_images[:BATCH_SIZE],y_:test_labels[:BATCH_SIZE]})
        print 'step:{}, train_accuracy={},test_accuracy={}'.format(i,train_accuracy,test_accuracy)
    sess.run(train_step,feed_dict={x:batch_xs,y_:batch_ys})

step:0, train_accuracy=0.10000000149,test_accuracy=0.10000000149
step:200, train_accuracy=0.879999995232,test_accuracy=0.759999990463
step:400, train_accuracy=0.920000016689,test_accuracy=0.77999997139
step:600, train_accuracy=0.899999976158,test_accuracy=0.759999990463
epoches copleted!
step:800, train_accuracy=0.879999995232,test_accuracy=0.77999997139
step:1000, train_accuracy=0.879999995232,test_accuracy=0.800000011921
step:1200, train_accuracy=0.879999995232,test_accuracy=0.819999992847
step:1400, train_accuracy=0.860000014305,test_accuracy=0.800000011921
epoches copleted!
step:1600, train_accuracy=0.879999995232,test_accuracy=0.819999992847
step:1800, train_accuracy=0.819999992847,test_accuracy=0.800000011921
step:2000, train_accuracy=0.899999976158,test_accuracy=0.819999992847
step:2200, train_accuracy=0.899999976158,test_accuracy=0.77999997139
epoches copleted!
step:2400, train_accuracy=0.959999978542,test_accuracy=0.800000011921
step:2600, train_accuracy=0.860000014305,test_accuracy=0.860000014305
step:2800, train_accuracy=0.959999978542,test_accuracy=0.819999992847
In [16]:
test_accuracy=accuracy.eval({x:test_images,y_:test_labels})
test_accuracy
Out[16]:
0.90450001
In [17]:
test_df.head()
Out[17]:
pixel0 pixel1 pixel2 pixel3 pixel4 pixel5 pixel6 pixel7 pixel8 pixel9 ... pixel774 pixel775 pixel776 pixel777 pixel778 pixel779 pixel780 pixel781 pixel782 pixel783
0 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
2 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0

5 rows × 784 columns

In [18]:
test_images=test_df.values
test_images=np.multiply(test_images,1.0/255.0)
predict = tf.argmax(y,1)
predicted_labels=predict.eval(feed_dict={x:test_images})
predicted_labels[:5]
Out[18]:
array([2, 0, 9, 9, 3])
In [19]:
submissions=pd.DataFrame({"ImageId": list(range(1,len(predicted_labels)+1)),
                         "Label": predicted_labels})
submissions.to_csv('submission_2.csv',index=False)