TensorFlow defines variables for "learning parameters"

The definition of learning parameters is very similar to the definition of input, which is divided into two parts: direct definition and dictionary definition.

Both of these are common usages, but the second case is commonly used in deep neural networks due to too many parameters.

a direct definition

1 Description

Parameters can be defined directly through tf.Variable.

2 Examples

# model parameters
W = tf.Variable(tf.random_normal([1]), name="weight")
b = tf.Variable(tf.zeros([1]), name="bias")

two dictionary definitions

1 Description

The dictionary definition is similar to the direct definition, but it is stacked together.

2 key codes

# model parameters
paradict = {
    'w': tf.Variable(tf.random_normal([1])),
    'b': tf.Variable(tf.zeros([1]))
}
# forward structure
z = tf.multiply(X, paradict['w'])+ paradict['b']

3 All codes

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
plotdata = { "batchsize":[], "loss":[] }
def moving_average(a, w=10):
    if len(a) < w:
        return a[:]    
    return [val if idx < w else sum(a[(idx-w):idx])/w for idx, val in enumerate(a)]
#generate simulation data
train_X = np.linspace(-1, 1, 100)
train_Y = 2 * train_X + np.random.randn(*train_X.shape) * 0.3 # y=2x, but with noise added
#Graphic display
plt.plot(train_X, train_Y, 'ro', label='Original data')
plt.legend()
plt.show()
# create model
# Placeholder
X = tf.placeholder("float")
Y = tf.placeholder("float")
# model parameters
paradict = {
    'w': tf.Variable(tf.random_normal([1])),
    'b': tf.Variable(tf.zeros([1]))
}
# forward structure
z = tf.multiply(X, paradict['w'])+ paradict['b']
#reverse optimization
cost =tf.reduce_mean( tf.square(Y - z))
learning_rate = 0.01
optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost) #Gradient descent
# Initialize variables
init = tf.global_variables_initializer()
#parameter settings
training_epochs = 20
display_step = 2
# start session
with tf.Session() as sess:
    sess.run(init)
    # Fit all training data
    for epoch in range(training_epochs):
        for (x, y) in zip(train_X, train_Y):
            sess.run(optimizer, feed_dict={X: x, Y: y})
        #Display training details
        if epoch % display_step == 0:
            loss = sess.run(cost, feed_dict={X: train_X, Y:train_Y})
            print ("Epoch:", epoch+1, "cost=", loss,"W=", sess.run(paradict['w']), "b=", sess.run(paradict['b']))
            if not (loss == "NA" ):
                plotdata["batchsize"].append(epoch)
                plotdata["loss"].append(loss)
    print (" Finished!")
    print ("cost=", sess.run(cost, feed_dict={X: train_X, Y: train_Y}), "W=", sess.run(paradict['w']), "b=", sess.run(paradict['b']))
    #Graphic display
    plt.plot(train_X, train_Y, 'ro', label='Original data')
    plt.plot(train_X, sess.run(paradict['w']) * train_X + sess.run(paradict['b']), label='Fitted line')
    plt.legend()
    plt.show()
    
    plotdata["avgloss"] = moving_average(plotdata["loss"])
    plt.figure(1)
    plt.subplot(211)
    plt.plot(plotdata["batchsize"], plotdata["avgloss"], 'b--')
    plt.xlabel('Minibatch number')
    plt.ylabel('Loss')
    plt.title('Minibatch run vs. Training loss')
     
    plt.show()
    print ("x=0.2，z=", sess.run(z, feed_dict={X: 0.2}))

4 Running results