003. evaluate and save model
<![CDATA[
# You inspected how to train model in previous lecture
# In this lecture,you will complete step of building model,
# and up to step of testing model
# If you input [1,5,3,7,8,10,12],
# and model ouputs [0,0,0,1,0],
# you can say model can predict well
import tensorflow as tf
input_data=[[1,5,3,7,8,10,12],
[5,8,10,3,9,7,1]]
label_data=[[0,0,0,1,0],
[1,0,0,0,0]]
INPUT_SIZE=7
HIDDEN1_SIZE=10
HIDDEN2_SIZE=8
CLASSES=5
Learning_rate=0.05
x_placeholder_node=tf.placeholder(tf.float32,shape=[None,INPUT_SIZE])
y_placeholder_node=tf.placeholder(tf.float32,shape=[None,CLASSES])
map_placeholder_and_data={x_placeholder_node:input_data,y_placeholder_node:label_data}
W_h1=tf.Variable(tf.truncated_normal(shape=[INPUT_SIZE,HIDDEN1_SIZE]),dtype=tf.float32)
b_h1=tf.Variable(tf.zeros(shape=[HIDDEN1_SIZE]),dtype=tf.float32)
W_h2=tf.Variable(tf.truncated_normal(shape=[HIDDEN1_SIZE,HIDDEN2_SIZE]),dtype=tf.float32)
b_h2=tf.Variable(tf.zeros(shape=[HIDDEN2_SIZE]),dtype=tf.float32)
W_o=tf.Variable(tf.truncated_normal(shape=[HIDDEN2_SIZE,CLASSES]),dtype=tf.float32)
b_o=tf.Variable(tf.zeros(shape=[CLASSES]),dtype=tf.float32)
# At point of finishing configuring weights,
# you create saver_object
# You can specify parameters you want to save
# When you don't specify parameters,
# all parameters you use will be saved
parameters_to_be_saved_list=[W_h1,b_h1,W_h2,b_h2,W_o,b_o]
saver_object=tf.train.Saver()
hidden1=tf.sigmoid(tf.matmul(x_placeholder_node,W_h1) + b_h1)
hidden2=tf.sigmoid(tf.matmul(hidden1,W_h2) + b_h2)
hypothesis_function=tf.sigmoid(tf.matmul(hidden2,W_o) + b_o)
cost=tf.reduce_sum(-y_placeholder_node*tf.log(hypothesis_function)-(1-y_placeholder_node)*tf.log(1-hypothesis_function),reduction_indices=1)
loss=tf.reduce_mean(cost)
train=tf.train.GradientDescentOptimizer(Learning_rate).minimize(cost)
# argmax() returns index of maximal value
# tf.argmax(input,dimension,name=None)
# You extract index of maximal value from prediction by using argmax()
# Then,you compare extracted index and index of label by using tf.equal()
# Then you can get like this [1,0,1,1]
# You calculate [1,0,1,1] to find mean value(3/4)
# For example,you will denote maximal value as 1
# index prediction label
# 0 [0 0 1 0 0] [0 0 1 0 0] => true => 1
# 1 [1 0 0 0 0] [0 1 0 0 0] => false => 0
# 2 [0 0 0 0 1] [0 0 0 0 1] => true => 1
# 3 [0 0 1 0 0] [0 0 1 0 0] => true => 1
# This code displays accuracy
# Since hypothesis_function is vector,you pass 1
comp_pred=tf.equal(tf.argmax(hypothesis_function,1) tf.argmax(y_placeholder_node,1))
accuracy=tf.reduce_mean(tf.cast(comp_pred,tf.float32))
sess_object=tf.Session()
sess_object.run(tf.initialize_all_variables())
for i in range(1000):
_,loss,acc=sess_object.run([train,cost,accuracy],map_placeholder_and_data)
pred=sess_object.run(tf.argmax(hypothesis_function,1),map_placeholder_and_data)
if i%100==0:
saver_object.save(sess_object,'./tflectcheckpoint')
print('----------')
print('step: ',i)
print('loss: ',loss)
print('accuracy: ',acc)
]]>