普洱信息港

当前位置:

如何利用微信监管你的TF训练

2019/08/16 来源:普洱信息港

导读

如何利用监管你的TF训练?按:本文作者Coldwings,本文整理自作者在知乎发布的文章《利用监管你的TF训练》,(公众号:)获其授权发

  如何利用监管你的TF训练?

  按:本文作者Coldwings,本文整理自作者在知乎发布的文章《利用监管你的TF训练》,(公众号:)获其授权发布。

  之前回答问题【在机器学习模型的训练期间,大概几十分钟到几小时不等,大家都会在等实验的时候做什么?】的时候,说到可以用来管着训练,完全不用守着。没想到这么受欢迎……

  原问题下的回答如下不知道有哪些朋友是在TF/keras/chainer/mxnet等框架下用python撸的….…

  这可是python啊……上itchat,弄个号加自己为好友(或者自己发自己),训练进展跟着一路发消息给自己就好了,做了可视化的话顺便把图也一并发过来

  。

  然后就能安心睡觉/逛街/泡妞/写答案了。

  讲道理,甚至简单的参数调整都可以照着用来……

  大体效果如下

  当然可以做得更全面一些。可靠的办法自然是干脆地做一个http服务或者一个rpc,然而这样往往太麻烦。本着简单高效的原则,几行代码能起到效果方便自己当然是的,接入或者web真就是不错的选择了。只是查看的话,TensorBoard就很好,但是如果想加入一些自定义操作,还是自行定制的。做成web,或者itchat做个服务,都是挺不赖的选择。

  正文如下

  这里折腾一个例子。以TensorFlow的example中,利用CNN处理MNIST的程序为例,我们做一点点小小的修改。

  首先这里放上写完的代码:

  #!/usr/bin/env python

  # coding: utf-8

  A Convolutional Network implementation example using TensorFlow library.

  This example is using the MNIST database of handwritten digits

  (

  Author: Aymeric Damien

  Project:

  Add a itchat controller with multi thread

  from __future__ import print_function

  import tensorflow as tf

  # Import MNIST data

  from ist import input_data

  # Import itchat threading

  import itchat

  import threading

  # Create a running status flag

  lock = ck()

  running = False

  # Parameters

  learning_rate = 0.001

  training_iters = 200000

  batch_size = 128

  display_step = 10

  def nn_train(wechat_name, param):

  global lock, running

  # Lock

  with lock:

  running = True

  # mnist data reading

  mnist = input_ad_data_sets(data/, one_hot=True)

  # Parameters

  # learning_rate = 0.001

  # training_iters = 200000

  # batch_size = 128

  # display_step = 10

  learning_rate, training_iters, batch_size, display_step = param

  # Network Parameters

  n_input = 784 # MNIST data input (img shape: 28*28)

  n_classes = 10 # MNIST total classes ( digits)

  dropout = 0.75 # Dropout, probability to keep units

  # tf Graph input

  x = aceholder(oat32, [None, n_input])

  y = aceholder(oat32, [None, n_classes])

  keep_prob = aceholder(oat32) #dropout (keep probability)

  # Create some wrappers for simplicity

  def conv2d(x, W, b, strides=1):

  # Conv2D wrapper, with bias and relu activation

  x = nv2d(x, W, strides=[1, strides, strides, 1], padding=SAME)

  x = as_add(x, b)

  return lu(x)

  def maxpool2d(x, k=2):

  # MaxPool2D wrapper

  return x_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1],

  padding=SAME)

  # Create model

  def conv_net(x, weights, biases, dropout):

  # Reshape input picture

  x = shape(x, shape=[-1, 28, 28, 1])

  # Convolution Layer

  conv1 = conv2d(x, weights[wc1], biases[bc1])

  # Max Pooling (down-sampling)

  conv1 = maxpool2d(conv1, k=2)

  # Convolution Layer

  conv2 = conv2d(conv1, weights[wc2], biases[bc2])

  # Max Pooling (down-sampling)

  conv2 = maxpool2d(conv2, k=2)

  # Fully connected layer

  # Reshape conv2 output to fit fully connected layer input

  fc1 = shape(conv2, [-1, weights[wd1].get_shape().as_list()[0]])

  fc1 = d(tmul(fc1, weights[wd1]), biases[bd1])

  fc1 = lu(fc1)

  # Apply Dropout

  fc1 = opout(fc1, dropout)

  # Output, class prediction

  out = d(tmul(fc1, weights[out]), biases[out])

  return out

  # Store layers weight bias

  weights = {

  # 5x5 conv, 1 input, 32 outputs

  wc1: riable(ndom_normal([5, 5, 1, 32])),

  # 5x5 conv, 32 inputs, 64 outputs

  wc2: riable(ndom_normal([5, 5, 32, 64])),

  # fully connected, 7*7*64 inputs, 1024 outputs

  wd1: riable(ndom_normal([7*7*64, 1024])),

  # 1024 inputs, 10 outputs (class prediction)

  out: riable(ndom_normal([1024, n_classes]))

  }

  biases = {

  bc1: riable(ndom_normal([32])),

  bc2: riable(ndom_normal([64])),

  bd1: riable(ndom_normal([1024])),

  out: riable(ndom_normal([n_classes]))

  }

  # Construct model

  pred = conv_net(x, weights, biases, keep_prob)

  # Define loss and optimizer

  cost = duce_mean(ftmax_cross_entropy_with_logits(logits=pred, labels=y))

  optimizer = amOptimizer(learning_rate=learning_rate).minimize(cost)

  # Evaluate model

  correct_pred = ual(gmax(pred, 1), gmax(y, 1))

  accuracy = duce_mean(st(correct_pred, oat32))

  # Initializing the variables

  init = obal_variables_initializer()

  # Launch the graph

  with ssion() as sess:

  n(init)

  step = 1

  # Keep training until reach max iterations

  print(Wait for lock)

  with lock:

  run_state = running

  print(Start)

  while step * batch_size training_iters and run_state:

  batch_x, batch_y = xt_batch(batch_size)

  # Run optimization op (backprop)

  n(optimizer, feed_dict={x: batch_x, y: batch_y,

  keep_prob: dropout})

  if step % display_step == 0:

  # Calculate batch loss and accuracy

  loss, acc = n([cost, accuracy], feed_dict={x: batch_x,

  y: batch_y,

  keep_prob: 1.})

  print(Iter + str(step*batch_size) + , Minibatch Loss= + \

  {:.6f}.format(loss) + , Training Accuracy= + \

  {:.5f}.format(acc))

  nd(Iter + str(step*batch_size) + , Minibatch Loss= + \

  {:.6f}.format(loss) + , Training Accuracy= + \

  {:.5f}.format(acc), wechat_name)

  step += 1

  with lock:

  run_state = running

  print(Optimization Finished!)

  nd(Optimization Finished!, wechat_name)

  # Calculate accuracy for 256 mnist test images

  print(Testing Accuracy:, \

  n(accuracy, feed_dict={x: ages[:256],

  y: bels[:256],

  keep_prob: 1.}))

  nd(Testing Accuracy: %s %

  n(accuracy, feed_dict={x: ages[:256],

  y: bels[:256],

  keep_prob: 1.}), wechat_name)

  with lock:

  running = False

  @g_register([XT])

  def chat_trigger(msg):

  global lock, running, learning_rate, training_iters, batch_size, display_step

  if msg[Text] == u开始:

  print(Starting)

  with lock:

  run_state = running

  if not run_state:

  try:

  read(target=nn_train, args=(msg[FromUserName], (learning_rate, training_iters, batch_size, display_step))).start()

  except:

  ply(Running)

  elif msg[Text] == u停止:

  print(Stopping)

  with lock:

  running = False

  elif msg[Text] == u参数:

  nd(lr=%f, ti=%d, bs=%d, ds=%d%(learning_rate, training_iters, batch_size, display_step),msg[FromUserName])

  else:

  try:

  param = msg[Text].split()

  key, value = param

  print(key, value)

  if key == lr:

  learning_rate = float(value)

  elif key == ti:

  training_iters = int(value)

  elif key == bs:

  batch_size = int(value)

  elif key == ds:

  display_step = int(value)

  except:

  pass

  if __name__ == __main__:

  to_login(hotReload=True)

  n()

  这段代码里面,我所做的修改主要是:

  0.导入了itchat和threading

  1. 把原本的脚本里络构成和训练的部分甩到了一个函数nn_train里

  def nn_train(wechat_name, param):

  global lock, running

  # Lock

  with lock:

  running = True

  # mnist data reading

  mnist = input_ad_data_sets(data/, one_hot=True)

  # Parameters

  # learning_rate = 0.001

  # training_iters = 200000

  # batch_size = 128

  # display_step = 10

  learning_rate, training_iters, batch_size, display_step = param

  # Network Parameters

  n_input = 784 # MNIST data input (img shape: 28*28)

  n_classes = 10 # MNIST total classes ( digits)

  dropout = 0.75 # Dropout, probability to keep units

  # tf Graph input

  x = aceholder(oat32, [None, n_input])

  y = aceholder(oat32, [None, n_classes])

  keep_prob = aceholder(oat32) #dropout (keep probability)

  # Create some wrappers for simplicity

  def conv2d(x, W, b, strides=1):

  # Conv2D wrapper, with bias and relu activation

  x = nv2d(x, W, strides=[1, strides, strides, 1], padding=SAME)

  x = as_add(x, b)

  return lu(x)

  def maxpool2d(x, k=2):

  # MaxPool2D wrapper

  return x_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1],

  padding=SAME)

  # Create model

  def conv_net(x, weights, biases, dropout):

  # Reshape input picture

  x = shape(x, shape=[-1, 28, 28, 1])

  # Convolution Layer

  conv1 = conv2d(x, weights[wc1], biases[bc1])

  # Max Pooling (down-sampling)

  conv1 = maxpool2d(conv1, k=2)

  # Convolution Layer

  conv2 = conv2d(conv1, weights[wc2], biases[bc2])

  # Max Pooling (down-sampling)

  conv2 = maxpool2d(conv2, k=2)

  # Fully connected layer

  # Reshape conv2 output to fit fully connected layer input

  fc1 = shape(conv2, [-1, weights[wd1].get_shape().as_list()[0]])

  fc1 = d(tmul(fc1, weights[wd1]), biases[bd1])

  fc1 = lu(fc1)

  # Apply Dropout

  fc1 = opout(fc1, dropout)

  # Output, class prediction

  out = d(tmul(fc1, weights[out]), biases[out])

  return out

  # Store layers weight bias

  weights = {

  # 5x5 conv, 1 input, 32 outputs

  wc1: riable(ndom_normal([5, 5, 1, 32])),

  # 5x5 conv, 32 inputs, 64 outputs

  wc2: riable(ndom_normal([5, 5, 32, 64])),

  # fully connected, 7*7*64 inputs, 1024 outputs

  wd1: riable(ndom_normal([7*7*64, 1024])),

  # 1024 inputs, 10 outputs (class prediction)

  out: riable(ndom_normal([1024, n_classes]))

  }

  biases = {

  bc1: riable(ndom_normal([32])),

  bc2: riable(ndom_normal([64])),

  bd1: riable(ndom_normal([1024])),

  out: riable(ndom_normal([n_classes]))

  }

  # Construct model

  pred = conv_net(x, weights, biases, keep_prob)

  # Define loss and optimizer

  cost = duce_mean(ftmax_cross_entropy_with_logits(logits=pred, labels=y))

  optimizer = amOptimizer(learning_rate=learning_rate).minimize(cost)

  # Evaluate model

  correct_pred = ual(gmax(pred, 1), gmax(y, 1))

  accuracy = duce_mean(st(correct_pred, oat32))

  # Initializing the variables

  init = obal_variables_initializer()

  # Launch the graph

  with ssion() as sess:

  n(init)

  step = 1

  # Keep training until reach max iterations

  print(Wait for lock)

  with lock:

  run_state = running

  print(Start)

  while step * batch_size training_iters and run_state:

  batch_x, batch_y = xt_batch(batch_size)

  # Run optimization op (backprop)

  n(optimizer, feed_dict={x: batch_x, y: batch_y,

  keep_prob: dropout})

  if step % display_step == 0:

  # Calculate batch loss and accuracy

  loss, acc = n([cost, accuracy], feed_dict={x: batch_x,

  y: batch_y,

  keep_prob: 1.})

  print(Iter + str(step*batch_size) + , Minibatch Loss= + \

  {:.6f}.format(loss) + , Training Accuracy= + \

  {:.5f}.format(acc))

  nd(Iter + str(step*batch_size) + , Minibatch Loss= + \

  {:.6f}.format(loss) + , Training Accuracy= + \

  {:.5f}.format(acc), wechat_name)

  step += 1

  with lock:

  run_state = running

  print(Optimization Finished!)

  nd(Optimization Finished!, wechat_name)

  # Calculate accuracy for 256 mnist test images

  print(Testing Accuracy:, \

  n(accuracy, feed_dict={x: ages[:256],

  y: bels[:256],

  keep_prob: 1.}))

  nd(Testing Accuracy: %s %

  n(accuracy, feed_dict={x: ages[:256],

  y: bels[:256],

  keep_prob: 1.}), wechat_name)

  with lock:

  running = False

  这里大部分是跟原本的代码一样的,不过首先所有print的地方都加了个nd来输出日志,此外加了个带锁的状态量running用来做运行开关。此外,部分参数是通过函数参数传入的。

  然后呢,写了个itchat的handler

  @g_register([XT])

  def chat_trigger(msg):

  global lock, running, learning_rate, training_iters, batch_size, display_step

  if msg[Text] == u开始:

  print(Starting)

  with lock:

  run_state = running

  if not run_state:

  try:

  read(target=nn_train, args=(msg[FromUserName], (learning_rate, training_iters, batch_size, display_step))).start()

  except:

  ply(Running)

  作用是,如果收到消息,内容为『开始』,那就跑训练的函数(当然,为了防止阻塞,放在了另一个线程里)

  再在脚本主流程里使用itchat登录并且启动itchat的服务,这样就实现了基本的控制。

  if __name__ == __main__:

  to_login(hotReload=True)

  n()

  但是我们不满足于此,我还希望可以对流程进行一些控制,对参数进行一些修改,于是乎:

  @g_register([XT])

  def chat_trigger(msg):

  global lock, running, learning_rate, training_iters, batch_size, display_step

  if msg[Text] == u开始:

  print(Starting)

  with lock:

  run_state = running

  if not run_state:

  try:

  read(target=nn_train, args=(msg[FromUserName], (learning_rate, training_iters, batch_size, display_step))).start()

  except:

  ply(Running)

  elif msg[Text] == u停止:

  print(Stopping)

  with lock:

  running = False

  elif msg[Text] == u参数:

  nd(lr=%f, ti=%d, bs=%d, ds=%d%(learning_rate, training_iters, batch_size, display_step),msg[FromUserName])

  else:

  try:

  param = msg[Text].split()

  key, value = param

  print(key, value)

  if key == lr:

  learning_rate = float(value)

  elif key == ti:

  training_iters = int(value)

  elif key == bs:

  batch_size = int(value)

  elif key == ds:

  display_step = int(value)

  except:

  pass

  通过这个,我们可以在epoch中途停止(因为nn_train里通过检查running标志来确定是否需要停下来),也可以在训练开始前调整learning_rate等几个参数。

  实在是很简单……

  版权文章,未经授权禁止转载。详情见转载须知。

小孩脸色发黄
小儿厌食的各种表现
胃肠敏感是什么意思
标签

友情链接