API tools faq
paste
Guest User

Untitled

a guest
Aug 15th, 2018
109
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 10.98 KB | None | 0 0
raw download clone embed print report
  1. {
  2. "cells": [
  3. {
  4. "cell_type": "markdown",
  5. "metadata": {},
  6. "source": [
  7. "I'd like to train VGG network as example, but it just ran too slow on my computer.\n",
  8. "\n",
  9. "I have only mac, so I can only run tensorflow with CPU :(\n",
  10. "\n",
  11. "So, I refer to our good old friend, handwriting recognition with MNIST.\n",
  12. "\n",
  13. "Bear with me, plz. I just want to show you a full process converting a graph to a single reusable file."
  14. ]
  15. },
  16. {
  17. "cell_type": "code",
  18. "execution_count": 1,
  19. "metadata": {
  20. "collapsed": true
  21. },
  22. "outputs": [],
  23. "source": [
  24. "from math import sqrt\n",
  25. "import numpy as np\n",
  26. "import tensorflow as tf\n",
  27. "from tensorflow.examples.tutorials.mnist.input_data import read_data_sets\n",
  28. "\n",
  29. "%matplotlib inline\n",
  30. "import matplotlib.pyplot as plt"
  31. ]
  32. },
  33. {
  34. "cell_type": "code",
  35. "execution_count": 2,
  36. "metadata": {
  37. "collapsed": true
  38. },
  39. "outputs": [],
  40. "source": [
  41. "# helper function\n",
  42. "def accuracy(y, y_):\n",
  43. " return (y == y_).mean()"
  44. ]
  45. },
  46. {
  47. "cell_type": "code",
  48. "execution_count": 3,
  49. "metadata": {},
  50. "outputs": [
  51. {
  52. "name": "stdout",
  53. "output_type": "stream",
  54. "text": [
  55. "Extracting mnist_data/train-images-idx3-ubyte.gz\n",
  56. "Extracting mnist_data/train-labels-idx1-ubyte.gz\n",
  57. "Extracting mnist_data/t10k-images-idx3-ubyte.gz\n",
  58. "Extracting mnist_data/t10k-labels-idx1-ubyte.gz\n"
  59. ]
  60. }
  61. ],
  62. "source": [
  63. "mnist = read_data_sets(\"mnist_data\", one_hot=True)"
  64. ]
  65. },
  66. {
  67. "cell_type": "code",
  68. "execution_count": 4,
  69. "metadata": {},
  70. "outputs": [],
  71. "source": [
  72. "# build a trainable graph\n",
  73. "train_graph = tf.Graph()\n",
  74. "hidden_shape = [100, 100]\n",
  75. "learning_rate = 0.05\n",
  76. "\n",
  77. "with train_graph.as_default():\n",
  78. " with tf.name_scope(\"input\"):\n",
  79. " image_batch = tf.placeholder(tf.float32, \n",
  80. " shape=[None, 784],\n",
  81. " name=\"image_batch\")\n",
  82. " layer_shapes = [784] + hidden_shape + [10]\n",
  83. " last_layer = image_batch\n",
  84. " with tf.name_scope(\"hidden\"):\n",
  85. " for in_shape, out_shape in zip(layer_shapes[:-1], layer_shapes[1:]):\n",
  86. " sig = sqrt(6)/(in_shape + out_shape)\n",
  87. " W_init = 2*sig*np.random.rand(in_shape, out_shape)-sig\n",
  88. " W = tf.Variable(np.random.randn(in_shape, out_shape),\n",
  89. " dtype=tf.float32,\n",
  90. " name=\"weight\")\n",
  91. " bias = tf.Variable(np.random.randn(out_shape),\n",
  92. " dtype=tf.float32,\n",
  93. " name=\"bias\")\n",
  94. " zscore = tf.matmul(last_layer, W) + bias\n",
  95. " last_layer = tf.nn.sigmoid(zscore)\n",
  96. " \n",
  97. " with tf.name_scope(\"output\"):\n",
  98. " prob = tf.nn.softmax(zscore, \n",
  99. " name=\"probability\")\n",
  100. " predict = tf.arg_max(prob, \n",
  101. " 1, \n",
  102. " name=\"prediction\")\n",
  103. " target = tf.placeholder(tf.float32,\n",
  104. " shape=[None, 10],\n",
  105. " name=\"target\")\n",
  106. " loss = -tf.reduce_mean(target*tf.log(prob))\n",
  107. " train_op = tf.train.AdagradOptimizer(learning_rate).minimize(loss)\n",
  108. " saver = tf.train.Saver()"
  109. ]
  110. },
  111. {
  112. "cell_type": "code",
  113. "execution_count": 5,
  114. "metadata": {},
  115. "outputs": [
  116. {
  117. "name": "stdout",
  118. "output_type": "stream",
  119. "text": [
  120. "Initialized\n",
  121. "Iteration 1000: 0.16 47.45%\n",
  122. "Iteration 2000: 0.12 62.55%\n",
  123. "Iteration 3000: 0.10 69.13%\n",
  124. "Iteration 4000: 0.08 73.12%\n",
  125. "Iteration 5000: 0.10 76.01%\n",
  126. "Iteration 6000: 0.06 77.92%\n",
  127. "Iteration 7000: 0.07 79.24%\n",
  128. "Iteration 8000: 0.06 80.53%\n",
  129. "Iteration 9000: 0.06 81.49%\n",
  130. "Iteration 10000: 0.07 82.52%\n",
  131. "Iteration 11000: 0.05 83.25%\n",
  132. "Iteration 12000: 0.06 83.75%\n",
  133. "Iteration 13000: 0.04 84.27%\n",
  134. "Iteration 14000: 0.06 84.71%\n",
  135. "Iteration 15000: 0.06 84.95%\n",
  136. "Iteration 16000: 0.05 85.33%\n",
  137. "Iteration 17000: 0.03 85.62%\n",
  138. "Iteration 18000: 0.05 86.05%\n",
  139. "Iteration 19000: 0.04 86.28%\n",
  140. "Iteration 20000: 0.05 86.50%\n"
  141. ]
  142. }
  143. ],
  144. "source": [
  145. "# Training and save\n",
  146. "num_iters = 20000\n",
  147. "batch_size = 300\n",
  148. "\n",
  149. "with tf.Session(graph=train_graph) as sess:\n",
  150. " tf.global_variables_initializer().run()\n",
  151. " print(\"Initialized\")\n",
  152. " \n",
  153. " for step in range(num_iters):\n",
  154. " images, labels = mnist.train.next_batch(batch_size)\n",
  155. " feed_dict = { image_batch: images,\n",
  156. " target: labels }\n",
  157. " l, _ = sess.run([loss, train_op], feed_dict=feed_dict)\n",
  158. " \n",
  159. " if (step+1) % 1000 == 0:\n",
  160. " pred = sess.run(predict, feed_dict={image_batch:mnist.test.images})\n",
  161. " acc = accuracy(pred, np.argmax(mnist.test.labels, axis=1))\n",
  162. " print(\"Iteration {}: {:.2f} {:.2f}%\".format(step+1, l, acc*100))\n",
  163. " \n",
  164. " saver.save(sess, save_path=\"model/mnist_example.chkp\", global_step=step)"
  165. ]
  166. },
  167. {
  168. "cell_type": "markdown",
  169. "metadata": {},
  170. "source": [
  171. "Looking good enough, ready to save the graph"
  172. ]
  173. },
  174. {
  175. "cell_type": "code",
  176. "execution_count": 6,
  177. "metadata": {},
  178. "outputs": [
  179. {
  180. "name": "stdout",
  181. "output_type": "stream",
  182. "text": [
  183. "checkpoint\r\n",
  184. "mnist_example.chkp-19999.data-00000-of-00001\r\n",
  185. "mnist_example.chkp-19999.index\r\n",
  186. "mnist_example.chkp-19999.meta\r\n"
  187. ]
  188. }
  189. ],
  190. "source": [
  191. "# saved data\n",
  192. "!ls model/"
  193. ]
  194. },
  195. {
  196. "cell_type": "code",
  197. "execution_count": 7,
  198. "metadata": {
  199. "collapsed": true
  200. },
  201. "outputs": [],
  202. "source": [
  203. "# import the graph_util\n",
  204. "# We'll use it to convert all variables in graph to constants\n",
  205. "from tensorflow.python.framework import graph_util"
  206. ]
  207. },
  208. {
  209. "cell_type": "code",
  210. "execution_count": 8,
  211. "metadata": {},
  212. "outputs": [
  213. {
  214. "data": {
  215. "text/plain": [
  216. "'model/mnist_example.chkp-19999'"
  217. ]
  218. },
  219. "execution_count": 8,
  220. "metadata": {},
  221. "output_type": "execute_result"
  222. }
  223. ],
  224. "source": [
  225. "# A CheckpointState object from tensorflow.python.training.checkpoint_state_pb2\n",
  226. "# Yes, it's produced by protobuf.\n",
  227. "checkpoint = tf.train.get_checkpoint_state(\"model/\")\n",
  228. "checkpoint.model_checkpoint_path"
  229. ]
  230. },
  231. {
  232. "cell_type": "code",
  233. "execution_count": 9,
  234. "metadata": {},
  235. "outputs": [
  236. {
  237. "data": {
  238. "text/plain": [
  239. "'model/mnist_example.chkp-19999.meta'"
  240. ]
  241. },
  242. "execution_count": 9,
  243. "metadata": {},
  244. "output_type": "execute_result"
  245. }
  246. ],
  247. "source": [
  248. "# meta file contains the meta data of the saved session\n",
  249. "meta_file_path = checkpoint.model_checkpoint_path+\".meta\"\n",
  250. "meta_file_path"
  251. ]
  252. },
  253. {
  254. "cell_type": "code",
  255. "execution_count": 10,
  256. "metadata": {
  257. "collapsed": true
  258. },
  259. "outputs": [],
  260. "source": [
  261. "# restore the saver from the meta file\n",
  262. "restore_graph = tf.Graph()\n",
  263. "with restore_graph.as_default():\n",
  264. " saver = tf.train.import_meta_graph(meta_file_path, clear_devices=True)\n",
  265. "restore_graph_def = restore_graph.as_graph_def()"
  266. ]
  267. },
  268. {
  269. "cell_type": "code",
  270. "execution_count": 11,
  271. "metadata": {},
  272. "outputs": [
  273. {
  274. "name": "stdout",
  275. "output_type": "stream",
  276. "text": [
  277. "INFO:tensorflow:Restoring parameters from model/mnist_example.chkp-19999\n",
  278. "INFO:tensorflow:Froze 6 variables.\n",
  279. "Converted 6 variables to const ops.\n"
  280. ]
  281. }
  282. ],
  283. "source": [
  284. "# restore the session with restored saver\n",
  285. "# convert graph to \"freezed\" graph\n",
  286. "# Note that you need to specify the layers you want to output, \n",
  287. "# tensorflow will extract a subgraph which involves the layers you want and freeze it.\n",
  288. "with tf.Session(graph=restore_graph) as sess:\n",
  289. " saver.restore(sess, checkpoint.model_checkpoint_path)\n",
  290. " out_graph_def = graph_util.convert_variables_to_constants(sess,\n",
  291. " restore_graph_def,\n",
  292. " [predict.op.name])"
  293. ]
  294. },
  295. {
  296. "cell_type": "code",
  297. "execution_count": 12,
  298. "metadata": {
  299. "collapsed": true
  300. },
  301. "outputs": [],
  302. "source": [
  303. "# write graph to disk just like what you will do with any protobuf object\n",
  304. "with tf.gfile.GFile(\"my_mnist.pb\", \"wb\") as fid:\n",
  305. " fid.write(out_graph_def.SerializeToString())"
  306. ]
  307. },
  308. {
  309. "cell_type": "markdown",
  310. "metadata": {},
  311. "source": [
  312. "------\n",
  313. "\n",
  314. "Testing time!"
  315. ]
  316. },
  317. {
  318. "cell_type": "code",
  319. "execution_count": 13,
  320. "metadata": {
  321. "collapsed": true
  322. },
  323. "outputs": [],
  324. "source": [
  325. "graph = tf.Graph()\n",
  326. "graph_def = graph.as_graph_def()"
  327. ]
  328. },
  329. {
  330. "cell_type": "code",
  331. "execution_count": 14,
  332. "metadata": {
  333. "collapsed": true
  334. },
  335. "outputs": [],
  336. "source": [
  337. "with tf.gfile.GFile(\"my_mnist.pb\", \"rb\") as rf:\n",
  338. " graph_def.ParseFromString(rf.read())\n",
  339. "with graph.as_default():\n",
  340. " tf.import_graph_def(graph_def, name=\"\")"
  341. ]
  342. },
  343. {
  344. "cell_type": "code",
  345. "execution_count": 15,
  346. "metadata": {},
  347. "outputs": [],
  348. "source": [
  349. "images_tensor = graph.get_tensor_by_name(\"input/image_batch:0\")\n",
  350. "predict_tensor = graph.get_tensor_by_name(\"output/prediction:0\")"
  351. ]
  352. },
  353. {
  354. "cell_type": "code",
  355. "execution_count": 16,
  356. "metadata": {},
  357. "outputs": [
  358. {
  359. "name": "stdout",
  360. "output_type": "stream",
  361. "text": [
  362. "86.5\n"
  363. ]
  364. }
  365. ],
  366. "source": [
  367. "with tf.Session(graph=graph) as sess:\n",
  368. " tf.global_variables_initializer().run()\n",
  369. " pred = sess.run(predict_tensor, \n",
  370. " feed_dict={images_tensor:mnist.test.images})\n",
  371. " print(accuracy(np.argmax(mnist.test.labels, axis=1), pred)*100)"
  372. ]
  373. },
  374. {
  375. "cell_type": "code",
  376. "execution_count": null,
  377. "metadata": {
  378. "collapsed": true
  379. },
  380. "outputs": [],
  381. "source": []
  382. }
  383. ],
  384. "metadata": {
  385. "kernelspec": {
  386. "display_name": "Python 3",
  387. "language": "python",
  388. "name": "python3"
  389. },
  390. "language_info": {
  391. "codemirror_mode": {
  392. "name": "ipython",
  393. "version": 3
  394. },
  395. "file_extension": ".py",
  396. "mimetype": "text/x-python",
  397. "name": "python",
  398. "nbconvert_exporter": "python",
  399. "pygments_lexer": "ipython3",
  400. "version": "3.6.1"
  401. }
  402. },
  403. "nbformat": 4,
  404. "nbformat_minor": 2
  405. }
Add Comment
Please, Sign In to add comment
Public Pastes
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!