#!/usr/bin/env python
# coding: utf-8

# In[1]:


import warnings
warnings.simplefilter(action='ignore')


# In[2]:


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


# ### 1. 以矩阵运算仿真神经网络

# #### 1.1 y = relu((X * W) + b)

# In[3]:


X = tf.Variable([[0.4, 0.2, 0.4]])
W = tf.Variable([
    [-0.5, -0.2],
    [-0.3, 0.4],
    [-0.5, 0.2]
])
b = tf.Variable([[0.1, 0.2]])
result = tf.matmul(X, W) + b
y = tf.nn.relu(result)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print('result:', sess.run(result))
    print('y:', sess.run(y))


# #### 1.2 y = sigmoid((X * W) + b)

# In[4]:


X = tf.Variable([[0.4, 0.2, 0.4]])
W = tf.Variable([
    [-0.5, -0.2],
    [-0.3, 0.4],
    [-0.5, 0.2]
])
b = tf.Variable([[0.1, 0.2]])
result = tf.matmul(X, W) + b
y = tf.nn.sigmoid(result)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print('result:', sess.run(result))
    print('y:', sess.run(y))


# #### 1.3 以正态分布的随机数生成权重与偏差的初始值

# In[5]:


X = tf.Variable([[0.4, 0.2, 0.4]])
W = tf.Variable(tf.random_normal([3, 2]))
b = tf.Variable(tf.random_normal([1, 2]))
result = tf.matmul(X, W) + b
y = tf.nn.relu(result)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    w_, b_, y_ = sess.run((W, b, y))
    print('W:', w_)
    print('b:', b_)
    print('y:', y_)


# #### 1.4 正态分布的随机数 tf.random_normal

# In[6]:


normal_array = tf.random_normal([1000])
with tf.Session() as sess:
    normal_data = sess.run(normal_array)

print('normal data length:', len(normal_data))
print(normal_data[:30])


# In[7]:


plt.hist(normal_data)
plt.show()


# ### 2. 以 placeholder 传入 X 值

# #### 2.1 以 placeholder 传入 1*3 的二维数组

# In[8]:


X = tf.placeholder('float', [None, 3])
W = tf.Variable(tf.random_normal([3, 2]))
b = tf.Variable(tf.random_normal([1, 2]))
result = tf.matmul(X, W) + b
y = tf.nn.relu(result)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    x_array = np.array([[0.4, 0.2, 0.4]])
    w_, b_, y_ = sess.run((W, b, y), feed_dict={X: x_array})
    print('W:', w_)
    print('b:', b_)
    print('y:', y_)


# #### 2.2 以 placeholder 传入 3*3 的二维数组

# In[9]:


X = tf.placeholder('float', [None, 3])
W = tf.Variable(tf.random_normal([3, 2]))
b = tf.Variable(tf.random_normal([1, 2]))
result = tf.matmul(X, W) + b
y = tf.nn.sigmoid(result)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    x_array = np.array([[0.4, 0.2, 0.4],
                        [0.3, 0.4, 0.5],
                        [0.3, -0.4, 0.5]])
    w_, b_, y_ = sess.run((W, b, y), feed_dict={X: x_array})
    print('W:', w_)
    print('b:', b_)
    print('y:', y_)


# ### 3. 创建 layer 函数以矩阵运算仿真神经网络

# #### 3.1 定义 layer 函数

# In[10]:


def layer(output_dim, input_dim, inputs, activation=None):
    W = tf.Variable(tf.random_normal([input_dim, output_dim]))
    b = tf.Variable(tf.random_normal([1, output_dim]))
    result = tf.matmul(inputs, W) + b
    
    if activation is None:
        outputs = result
    else:
        outputs = activation(result)
    
    return outputs


# In[11]:


X = tf.placeholder('float', [None, 4])
y = layer(output_dim=3, input_dim=4, inputs=X, activation=tf.nn.relu)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    x_array = np.array([[0.4, 0.2, 0.4, 0.1],
                        [0.3, 0.4, 0.5, 0.3],
                        [0.3, -0.4, 0.5, 0.2]])
    x_, y_ = sess.run((X, y), feed_dict={X: x_array})
    print('X:', x_)
    print('y:', y_)


# #### 3.2 使用 layer 函数建立3层神经网络

# In[12]:


X = tf.placeholder('float', [None, 4])
h = layer(output_dim=3, input_dim=4, inputs=X, activation=tf.nn.relu)
y = layer(output_dim=2, input_dim=3, inputs=h)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    x_array = np.array([[0.4, 0.2, 0.4, 0.5]])
    x_, h_, y_ = sess.run((X, h, y), feed_dict={X: x_array})
    print('X:', x_)
    print('h:', h_)
    print('y:', y_)


# ### 4. 创建 layer_debug 函数显示权重与偏差

# In[13]:


def layer_debug(output_dim, input_dim, inputs, activation=None):
    W = tf.Variable(tf.random_normal([input_dim, output_dim]))
    b = tf.Variable(tf.random_normal([1, output_dim]))
    result = tf.matmul(inputs, W) + b
    
    if activation is None:
        outputs = result
    else:
        outputs = activation(result)
    
    return outputs, W, b


# In[14]:


X = tf.placeholder('float', [None, 4])
h, W1, b1 = layer_debug(output_dim=3, input_dim=4, inputs=X, activation=tf.nn.relu)
y, W2, b2 = layer_debug(output_dim=2, input_dim=3, inputs=h)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    x_array = np.array([[0.4, 0.2, 0.4, 0.5]])
    x_, h_, y_, w1_, b1_, w2_, b2_ = sess.run((X, h, y, W1, b1, W2, b2), feed_dict={X: x_array})
    print('X:', x_)
    print('h:', h_)
    print('W1:', w1_)
    print('b1:', b1_)
    print('y:', y_)
    print('W2:', w2_)
    print('b2:', b2_)