import tensorflow as tf
"""
Mnist by tensorflow 
"""
# 通过Tensorflow自带模块，加载Mnist数据集 
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
# Tensorflow依赖于高效的C++后端来进行计算，与后端的连接叫做session，一般是先构建图，然后在session中启动它
# 更方便的是InteractiveSession类，可以在运行图的时候，动态构建图
sess = tf.InteractiveSession()
# 占位符
# 我们通过为输入图像和目标输出类别创建节点，来开始构建计算图。
x = tf.placeholder("float", shape=[None, 784])
y_ = tf.placeholder("float", shape=[None, 10])
# 变量 w 为一个[784.10]的矩阵，b是一个10维向量
w = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))
# 变量初始化
sess.run(tf.global_variables_initializer())
# 实现softmax回归模型
y = tf.nn.softmax(tf.matmul(x,w)+b)
# loss:整个Mini batch的交叉熵
cross_entropy = -tf.reduce_sum(y_*tf.log(y))
# BP算法
# 返回的train_step对象，运行时会使用梯度下降来更新参数，整个模型的训练通过反复运行train_step来进行
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)
# 开始反复训练
for i in range(1000):
    batch = mnist.train.next_batch(50)
    # feed_dict 用数据替换张量x, y_
    sess.run(train_step, feed_dict={x:batch[0], y_:batch[1]})

# 模型评估
prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
# 将上面返回的bool数组转换为float，用以计算准确率
accuracy = tf.reduce_mean(tf.cast(prediction, "float"))
Acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels})
print ("Accuracy: ", Acc)

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
sess = tf.InteractiveSession()
x = tf.placeholder("float", shape=[None, 784])
y_ = tf.placeholder("float", shape=[None, 10])
# CNN网络训练时需要随机初始化权重
# tf.truncated_normal(shape, mean, stddev)  shape表示生成张量的维度，mean是均值，stddev是标准差
# 截断正态分布函数，产生正态分布的值如果与均值的差值大于两倍的标准差，那就重新生成
def weight_variable(shape):
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial)
def bias_variable(shape):
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial) # 用0.1赋值的常量
def conv2d(x, W):
    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
def max_pool_2x2(x):
    return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')

# patch 大小5*5
W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])

x_image = tf.reshape(x, [-1,28,28,1])  # reshape_1，batch*W*H*C

h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)

W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])

h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)

W_fc1 = weight_variable([7 * 7 * 64, 1024])
b_fc1 = bias_variable([1024])

h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64]) # reshape_2 变为1维向量
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

keep_prob = tf.placeholder("float")
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob) # 添加一个占位符，在train、test中灵活改变keep_prob比例

W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])

y_conv=tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)

# 模型其他参数
cross_entropy = -tf.reduce_sum(y_*tf.log(y_conv))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
sess.run(tf.global_variables_initializer())
for i in range(20000):
    batch = mnist.train.next_batch(50)
    if i % 100 == 0:
        train_accuracy = sess.run(accuracy, feed_dict={x:batch[0], y_: batch[1], keep_prob: 1.0})
        print ("step %d, training accuracy %g" % (i, train_accuracy))
    train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5}) # 循环往复执行优化计算
Acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})
# 也可使用Tensor.eval() 和 Operation.run() 方法代替 Session.run()
# print (accuracy.run(feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))
print ("test accuracy", Acc)

class PsnrEstimator(object):
    """psnr评估器
    """
    def func(self):
        se = tf.square(self.tf_origin_image - self.tf_recon_image)
        mse = tf.reduce_mean(se) # 每个通道分别计算MSE取平均
        psnr = tf.cond(tf.equal(mse, 0),
                       lambda: tf.constant(100, dtype=tf.float32),
                       lambda: 10 * (tf.log(255 * 255 / mse) / np.log(10))) # 换底公式
        return se, psnr

ssh-keygen -t rsa -C "wuxiangji@tucodec.com" 
# 上述命令会在用户主目录的.ssh目录下，生成id_rsa(私钥)和id_rsa.pub(公钥)
# 登陆Git，将公钥添加到“Account settings”，“SSH Keys”即可
# Git的ssh key本质上就是建立了本机与服务器的连接，而且免去了以后输入密码的烦恼

# cd 项目跟目录
git init 					# 建立本地git仓库
git config user.name "gitlab's Name"
git config user.email "gitlab@xx.com"
git config --list # 查看当前配置, 在当前项目下面查看的配置是全局配置+当前项目的配置, 使用的时候会优先使用当前项目的配置

Host 192.28.1.81
    HostName 192.28.1.81
    Port 2222
    User Xiangji_WU
    IdentityFile ~/.ssh/id_rsa_xxx

Host 192.28.1.83
    HostName 192.28.1.83
    Port 2222
    User Xiangji_WU
    IdentityFile ~/.ssh/id_rsa_yyy

# cd 项目跟目录
git init # 建立本地git仓库
git add . 
git commit -m "first commit" 
git remote add origin git@github.com:Wxjwjj/Facial-Landmark.git # 关联某个远程库
git pull origin master 		# 要取回origin主机的分支，与本地的master分支合并
git push -u origin master  	# 第一次推送master分支的所有内容

git status 					# 查看当前的git仓库状态
git diff readme.txt 		# 查看修改后的readme.txt和之前的区别
git add readme.txt  		# 文件添加到暂存区 注意每次修改，如果不add到暂存区，那就不会加入到commit中
git commit -m "更新说明" 		# 暂存区的内容提交到当前master分支
git log --pretty=oneline 	# 查看一共提交了多少版本
git reflog 					# 查看命令历史
git reset --hard  head^ 
# HEAD指向的版本就是当前版本，当要在多个版本的历史之间穿梭，使用命令 git reset --hard commit_id
git reset HEAD readme.txt 	# 已经添加到暂存区的修改
git checkout -- readme.txt 	# 进撤销工作区内容的修改
git checkout -- readme.txt 	# rm指令误删的文件可以从版本库中进行恢复
git rm readme.txt 			# 从版本库中彻底删除，并且要进行commit
git commit -m "remove test.txt" 
git pull 					# 当前分支自动与唯一一个追踪分支进行合并
git push origin master

brew install openblas
brew install opencv

cd /usr/local/opt
ln -s /opt/X11 X11