tensorflow 手写数字识别

https://www.kaggle.com/kakauandme/tensorflow-deep-nn

本人只是负责将这个kernels的代码整理了一遍,具体还是请看原链接

import numpy as np
import pandas as pd
import tensorflow

# settings
LEARNING_RATE = 1e-4
# set to 20000 on local environment to get 0.99 accuracy
TRAINING_ITERATIONS = 20000
    
DROPOUT = 0.5
BATCH_SIZE = 50

# set to 0 to train on all available data
VALIDATION_SIZE = 2000

# image number to output
IMAGE_TO_DISPLAY = 10

# read training data from CSV file 
data = pd.read_csv('D://kaggle//DigitRecognizer//data//train.csv')

images = data.iloc[:,1:].values
images = images.astype(np.float)
# convert from [0:255] => [0.0:1.0]
images = np.multiply(images, 1.0 / 255.0)

image_size = images.shape[1]
print ('image_size => {0}'.format(image_size))

# in this case all images are square
image_width = image_height = np.ceil(np.sqrt(image_size)).astype(np.uint8)

print ('image_width => {0}\nimage_height => {1}'.format(image_width,image_height))

labels_flat = data.iloc[:,0].values

print('labels_flat({0})'.format(len(labels_flat)))
print ('labels_flat[{0}] => {1}'.format(IMAGE_TO_DISPLAY,labels_flat[IMAGE_TO_DISPLAY]))

labels_count = np.unique(labels_flat).shape[0]

print('labels_count => {0}'.format(labels_count))

def dense_to_one_hot(labels_dense, num_classes):
    num_labels = labels_dense.shape[0]
    index_offset = np.arange(num_labels) * num_classes
    labels_one_hot = np.zeros((num_labels, num_classes))
    labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1
    return labels_one_hot

labels = dense_to_one_hot(labels_flat, labels_count)
labels = labels.astype(np.uint8)

print('labels({0[0]},{0[1]})'.format(labels.shape))
print ('labels[{0}] => {1}'.format(IMAGE_TO_DISPLAY,labels[IMAGE_TO_DISPLAY]))

# split data into training & validation
validation_images = images[:VALIDATION_SIZE]
validation_labels = labels[:VALIDATION_SIZE]

train_images = images[VALIDATION_SIZE:]
train_labels = labels[VALIDATION_SIZE:]


print('train_images({0[0]},{0[1]})'.format(train_images.shape))
print('validation_images({0[0]},{0[1]})'.format(validation_images.shape))


# weight initialization
def weight_variable(shape):
    initial = tensorflow.truncated_normal(shape, stddev=0.1)
    return tensorflow.Variable(initial)

def bias_variable(shape):
    initial = tensorflow.constant(0.1, shape=shape)
    return tensorflow.Variable(initial)

# convolution
def conv2d(x, W):
    return tensorflow.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')

# pooling
# [[0,3],
#  [4,2]] => 4

# [[0,1],
#  [1,1]] => 1

def max_pool_2x2(x):
    return tensorflow.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')

# input & output of NN

# images
x = tensorflow.placeholder('float', shape=[None, image_size])
# labels
y_ = tensorflow.placeholder('float', shape=[None, labels_count])

# first convolutional layer
W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])

# (40000,784) => (40000,28,28,1)
image = tensorflow.reshape(x, [-1,image_width , image_height,1])
#print (image.get_shape()) # =>(40000,28,28,1)


h_conv1 = tensorflow.nn.relu(conv2d(image, W_conv1) + b_conv1)
#print (h_conv1.get_shape()) # => (40000, 28, 28, 32)
h_pool1 = max_pool_2x2(h_conv1)
#print (h_pool1.get_shape()) # => (40000, 14, 14, 32)


# Prepare for visualization
# display 32 fetures in 4 by 8 grid
layer1 = tensorflow.reshape(h_conv1, (-1, image_height, image_width, 4 ,8))  

# reorder so the channels are in the first dimension, x and y follow.
layer1 = tensorflow.transpose(layer1, (0, 3, 1, 4,2))

layer1 = tensorflow.reshape(layer1, (-1, image_height*4, image_width*8)) 

# second convolutional layer
W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])

h_conv2 = tensorflow.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
#print (h_conv2.get_shape()) # => (40000, 14,14, 64)
h_pool2 = max_pool_2x2(h_conv2)
#print (h_pool2.get_shape()) # => (40000, 7, 7, 64)

# Prepare for visualization
# display 64 fetures in 4 by 16 grid
layer2 = tensorflow.reshape(h_conv2, (-1, 14, 14, 4 ,16))  

# reorder so the channels are in the first dimension, x and y follow.
layer2 = tensorflow.transpose(layer2, (0, 3, 1, 4,2))

layer2 = tensorflow.reshape(layer2, (-1, 14*4, 14*16)) 


# densely connected layer
W_fc1 = weight_variable([7 * 7 * 64, 1024])
b_fc1 = bias_variable([1024])

# (40000, 7, 7, 64) => (40000, 3136)
h_pool2_flat = tensorflow.reshape(h_pool2, [-1, 7*7*64])

h_fc1 = tensorflow.nn.relu(tensorflow.matmul(h_pool2_flat, W_fc1) + b_fc1)
#print (h_fc1.get_shape()) # => (40000, 1024)

# dropout
keep_prob = tensorflow.placeholder('float')
h_fc1_drop = tensorflow.nn.dropout(h_fc1, keep_prob)


# readout layer for deep net
W_fc2 = weight_variable([1024, labels_count])
b_fc2 = bias_variable([labels_count])

y = tensorflow.nn.softmax(tensorflow.matmul(h_fc1_drop, W_fc2) + b_fc2)

#print (y.get_shape()) # => (40000, 10)


# cost function
cross_entropy = -tensorflow.reduce_sum(y_*tensorflow.log(y))


# optimisation function
train_step = tensorflow.train.AdamOptimizer(LEARNING_RATE).minimize(cross_entropy)

# evaluation
correct_prediction = tensorflow.equal(tensorflow.argmax(y,1),tensorflow.argmax(y_,1))

accuracy = tensorflow.reduce_mean(tensorflow.cast(correct_prediction, 'float'))

# prediction function
#[0.1, 0.9, 0.2, 0.1, 0.1 0.3, 0.5, 0.1, 0.2, 0.3] => 1
predict = tensorflow.argmax(y,1)

epochs_completed = 0
index_in_epoch = 0
num_examples = train_images.shape[0]

# serve data by batches
def next_batch(batch_size):
    
    global train_images
    global train_labels
    global index_in_epoch
    global epochs_completed
    
    start = index_in_epoch
    index_in_epoch += batch_size
    
    # when all trainig data have been already used, it is reorder randomly    
    if index_in_epoch > num_examples:
        # finished epoch
        epochs_completed += 1
        # shuffle the data
        perm = np.arange(num_examples)
        np.random.shuffle(perm)
        train_images = train_images[perm]
        train_labels = train_labels[perm]
        # start next 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]


# start TensorFlow session
init = tensorflow.initialize_all_variables()
sess = tensorflow.InteractiveSession()

sess.run(init)

# visualisation variables
train_accuracies = []
validation_accuracies = []
x_range = []

display_step=1

for i in range(TRAINING_ITERATIONS):

    #get new batch
    batch_xs, batch_ys = next_batch(BATCH_SIZE)        

    # check progress on every 1st,2nd,...,10th,20th,...,100th... step
    if i%display_step == 0 or (i+1) == TRAINING_ITERATIONS:
        
        train_accuracy = accuracy.eval(feed_dict={x:batch_xs, 
                                                y_: batch_ys, 
                                                keep_prob: 1.0})       
        if(VALIDATION_SIZE):
            validation_accuracy = accuracy.eval(feed_dict={ x: validation_images[0:BATCH_SIZE], 
                                                            y_: validation_labels[0:BATCH_SIZE], 
                                                            keep_prob: 1.0})                                  
            print('training_accuracy / validation_accuracy => %.2f / %.2f for step %d'%(train_accuracy, validation_accuracy, i))
            
            validation_accuracies.append(validation_accuracy)
            
        else:
            print('training_accuracy => %.4f for step %d'%(train_accuracy, i))
        train_accuracies.append(train_accuracy)
        x_range.append(i)
        
        # increase display_step
        if i%(display_step*10) == 0 and i:
            display_step *= 10
    # train on batch
    sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys, keep_prob: DROPOUT})

# read test data from CSV file 
test_images = pd.read_csv('D://kaggle//DigitRecognizer//data//test.csv').values
test_images = test_images.astype(np.float)

# convert from [0:255] => [0.0:1.0]
test_images = np.multiply(test_images, 1.0 / 255.0)

print('test_images({0[0]},{0[1]})'.format(test_images.shape))


# predict test set
#predicted_lables = predict.eval(feed_dict={x: test_images, keep_prob: 1.0})

# using batches is more resource efficient
predicted_lables = np.zeros(test_images.shape[0])
for i in range(0,test_images.shape[0]//BATCH_SIZE):
    predicted_lables[i*BATCH_SIZE : (i+1)*BATCH_SIZE] = predict.eval(feed_dict={x: test_images[i*BATCH_SIZE : (i+1)*BATCH_SIZE], 
                                                                                keep_prob: 1.0})


print('predicted_lables({0})'.format(len(predicted_lables)))

# output test image and prediction
#   display(test_images[IMAGE_TO_DISPLAY])
print ('predicted_lables[{0}] => {1}'.format(IMAGE_TO_DISPLAY,predicted_lables[IMAGE_TO_DISPLAY]))

# save results
np.savetxt('D://kaggle//DigitRecognizer//submission_softmax.csv', 
        np.c_[range(1,len(test_images)+1),predicted_lables], 
        delimiter=',', 
        header = 'ImageId,Label', 
        comments = '', 
        fmt='%d')
layer1_grid = layer1.eval(feed_dict={x: test_images[IMAGE_TO_DISPLAY:IMAGE_TO_DISPLAY+1], keep_prob: 1.0})
sess.close()
posted @ 2017-06-20 21:52 qscqesze 阅读(...) 评论(...) 编辑 收藏