Import tensorflow TF AS
Import numpy AS NP
# ## defined neural network layer functions added START ###
DEF add_layer (Inputs, in_size, out_size, activation_function = None):
"" " Description: Add neural network layer functions.
: Param inputs: an input neural layer
: Number of neurons in the input layer of neural: param in_size
: Number of neurons in the output layer neural: param out_size
: Param activation_function: activation function
"""
# Define a "in_size row, out_size column" random matrix variables
Weights = tf.Variable (tf.random_normal ([in_size, out_size]))
# Define a "row 1, out_size column" 0 reference variable value matrix
biases = tf.Variable (tf.zeros ([1, out_size]) + 0.1 )
# Define a matrix multiplication function formula
Wx_plus_b = tf.matmul (Inputs, Weights) + biases
# Determines whether the excitation function
IF activation_function IS None:
outputs=Wx_plus_b
else:
Outputs = activation_function (Wx_plus_b)
return Outputs
# ## defined neural network layer functions added END ###
# ## define a variable structure START ###
# Define the start input: Returns an array of 300 rows 300 of evenly spaced intervals specified -1 to 1, and then converted into an array of 300 rows of the matrix
# example:
# X1 = np.array ( [. 1, 2,. 3,. 4,. 5])
# # The Shape of X1 IS (. 5,)
# x1_new = X1 [:, np.newaxis]
# # now, The Shape of x1_new IS (. 5,. 1)
# Array ([[. 1],
# [2],
# [. 3],
# [. 4],
# [. 5]])
# x1_new = X1 [np.newaxis ,:]
# # now, The Shape of x1_new IS (. 1, . 5)
# Array ([[. 1, 2,. 3,. 4,. 5]])
x_data = np.linspace (-1,1,300 ) [:, np.newaxis]
# Define Noise: using a Gaussian distribution of probability density function is defined a mean of 0 and standard deviation for the Gaussian random number 0.05, the number is the number of matrix elements x_data
Noise np.random.normal = (0, 0.05 , x_data.shape)
# Define the start output: x_data subtracting the square of 0.5, plus noise
y_data = np.square (x_data) -0.5+ Noise
# Define variable runtime parameters
XS = tf.placeholder (tf.float32, [None,. 1 ])
ys=tf.placeholder(tf.float32,[None,1])
# ## is defined neural network architecture START ###
# Define hidden layer neural network layer Layer01
Layer01 add_layer = (XS, 1,10, activation_function = tf.nn.relu)
# define hidden layer neural network layer Layer02
Layer02 = add_layer (layer01,10,10, activation_function = tf.nn. Sigmoid)
# define the output layer prediction prediction
prediction = add_layer (layer02,10,1, activation_function = None)
# calculated loss
# 1 calculates a deviation of the predicted output and the output of the start square
loss_square = tf.square (y_data - prediction)
# 2. Double each dimension computing a sum amount of elemental
reduce_sum_square = tf.reduce_sum (loss_square, reduction_indices = [. 1 ])
# 3. calculated loss: on each dimension element tensor average
loss = tf.reduce_mean (reduce_sum_square)
# Using a gradient descent algorithm to train all samples
train_step = tf.train.GradientDescentOptimizer (0.1 ) .minimize (Loss)
# define a variable initialization
the init = tf.initialize_all_variables ()
# Create Session
sess = tf.Session ()
# run the initialization variable pointer
sess .run (init)
# ## is defined neural network END ###
# ## define the variables of Structural END ###
for i in range(2000):
sess.run(train_step,feed_dict={xs:x_data,ys:y_data})
if i%50==0:
print(sess.run(loss,feed_dict={xs:x_data,ys:y_data}))