使用TenforFlow 搭建BP神經網絡擬合二次函數

使用簡單BP神經網絡擬合二次函數

當擁有兩層神經元時候，擬合程度明顯比一層好
並出現以下警告：

C:\Program Files\Python36\lib\site-packages\matplotlib\backend_bases.py:2453: MatplotlibDeprecationWarning: Using default event loop until function specific to this GUI is implemented warnings.warn(str, mplDeprecation)

偶爾畫出直線，不知爲什麼
當Learning Rate越高，或者層數越多
畫出直線或者罷工的概率就越大，願評論區能出現解答
罷工的緣由： NaN
直線：未知？，NN學錯東西了

Code（nn with 2 hidden layer）

import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np

# refenrence: http://blog.csdn.net/jacke121/article/details/74938031

''' numpy.linspace test  
import numpy as np
# list_random1 = np.linspace(-1, 1, 300)
# list_random2 = np.linspace(-1, 1, 300)[:, np.newaxis]
# print(list_random1)
# print(list_random1)
# print(np.shape(list_random1), np.shape(list_random2))
#     (300)                   (300, 1)

x_data = np.linspace(-1,1,10)[:, np.newaxis]
noise = np.random.normal(0, 0.05, x_data.shape)
y_data = np.square(x_data) - 0.5 + noise
x_data2 = tf.random_uniform([10, 1], -1, 1)
y_data2 = tf.square(x_data2) - tf.random_normal(x_data2.shape, 2)
print(x_data, '\n\n', x_data.shape, '\n\n')     # (10, 1) 
print(y_data, '\n\n', y_data.shape, '\n\n')  
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(sess.run(x_data2))
    print(x_data2)                              # Tensor("random_uniform:0", shape=(10, 1), dtype=float32)
    print(sess.run(y_data2))
    print(y_data2)
'''



# add layer function
def add_layer(inputs, input_size, output_size, activation_function = None):
    # add one more layer and return the output of this layer
    Weights = tf.Variable(tf.random_normal([input_size, output_size]))
    biases = tf.Variable(tf.zeros([1, output_size]) + 0.1)
    outputs = tf.matmul(inputs, Weights) + biases
    if activation_function is not None:
        outputs = activation_function(outputs)
    return outputs



# creat data
# x_data: random[-1, 1) shape=[100,1] uniform distribute
# y_data: normal distribute(2) of random numbers, shape = [100, 1]
x_data = np.linspace(-1,1,100)[:, np.newaxis]
noise = np.random.normal(0, 0.05, x_data.shape)
y_data = np.square(x_data) - 0.5 + noise

# define placehold for inputs of nn
x = tf.placeholder(tf.float32, [None, 1])
y = tf.placeholder(tf.float32, [None, 1])

# hidden layer
layer1 = add_layer(x, 1, 10, activation_function=tf.nn.relu)
# hidden layer2
layer2 = add_layer(layer1, 10, 10, activation_function=tf.nn.relu)
# output layer
output_layer = add_layer(layer2, 10, 1, activation_function=None)

# define loss for nn
loss = tf.reduce_mean(tf.reduce_sum(tf.square(y - output_layer),
                     reduction_indices=[1]))
train = tf.train.GradientDescentOptimizer(0.1).minimize(loss)

# visualize the result
fig = plt.figure()
ax = fig.add_subplot(1,1,1)
ax.scatter(x_data, y_data)
plt.ion()
plt.show()

with tf.Session() as sess:
    sess.run(tf.initialize_all_variables())
    for i in range(1, 500):
        sess.run(train, feed_dict={x: x_data, y: y_data})
        # visualize the result
        if i%20 == 0:
            try:
                ax.lines.remove(lines[0])
            except Exception:
                pass
            output = sess.run(output_layer, feed_dict={x: x_data})
            lines = ax.plot(x_data, output, 'r-', lw=5)
            plt.pause(0.1)

plt.pause(100)