# coding:utf8
import numpy as np
import cPickle
import os
import tensorflow as tf
class SoftMax:
def __init__(self,MAXT=30,step=0.0025):
self.MAXT = MAXT
self.step = step
def load_theta(self,datapath="data/softmax.pkl"):
self.theta = cPickle.load(open(datapath,'rb'))
def process_train(self,data,label,typenum=10,batch_size=500):
batches = data.shape[0] / batch_size
valuenum=data.shape[1]
if len(label.shape)==1:
label=self.reshape_data(label,typenum)
x = tf.placeholder("float", [None,valuenum])
theta = tf.Variable(tf.zeros([valuenum,typenum]))
y = tf.nn.softmax(tf.matmul(x,theta))
y_ = tf.placeholder("float", [None, typenum])
cross_entropy = -tf.reduce_sum(y_*tf.log(y)) #交叉熵
train_step = tf.train.GradientDescentOptimizer(self.step).minimize(cross_entropy)
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
for epoch in range(self.MAXT):
cost_=[]
for index in xrange(batches):
c_,_=sess.run([cross_entropy,train_step], feed_dict={ x: data[index * batch_size: (index + 1) * batch_size],
y_: label[index * batch_size: (index + 1) * batch_size]})
cost_.append(c_)
if epoch % 5 == 0:
print(( 'epoch %i, minibatch %i/%i,averange cost is %f') %
(epoch,index + 1,batches,np.mean(cost_)))
self.theta=sess.run(theta)
if not os.path.exists('data/softmax.pkl'):
f= open("data/softmax.pkl",'wb')
cPickle.dump(self.theta,f)
f.close()
return self.theta
def process_test(self,data,label,typenum=10):
valuenum=data.shape[1]
if len(label.shape)==1:
label=self.reshape_data(label,typenum)
x = tf.placeholder("float", [None,valuenum])
theta = self.theta
y = tf.nn.softmax(tf.matmul(x,theta))
y_ = tf.placeholder("float", [None, typenum])
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print "Accuracy: ",sess.run(accuracy, feed_dict={x: data,y_: label})
def h(self,x):
m = np.exp(np.dot(x,self.theta))
sump = np.sum(m,axis=1)
return m/sump
def predict(self,x):
return np.argmax(self.h(x),axis=1)
def reshape_data(self,label,typenum):
label_=[]
for yl_ in label:
tl_=np.zeros(typenum)
tl_[yl_]=1.0
label_.append(tl_)
return np.mat(label_)
if __name__ == '__main__':
f = open('mnist.pkl', 'rb')
training_data, validation_data, test_data = cPickle.load(f)
training_inputs = [np.reshape(x, 784) for x in training_data[0]]
data = np.array(training_inputs)
training_inputs = [np.reshape(x, 784) for x in validation_data[0]]
vdata = np.array(training_inputs)
f.close()
softmax = SoftMax()
softmax.process_train(data,training_data[1])
softmax.process_test(vdata,validation_data[1]) #Accuracy: 0.9269
softmax.process_test(data,training_data[1]) #Accuracy: 0.92718
