utensil.tngl.sh
A series of experiments with Julia and Python
Try MXNet and MXBoard on Windows
author
utensil
date
(Jul 2, 2018, 11:16 PM +0800)
commit
0e3bb42c
0e3bb42c736eb56cd0942f2580248700e6f29a32
parent
b7183369
b71833698a8428c8443e9f887fcfd93a3194c773
+730
+730
dl/hello_mx_gpu.ipynb
+730
dl/hello_mx_gpu.ipynb
···
1
+
{
2
+
"cells": [
3
+
{
4
+
"cell_type": "markdown",
5
+
"metadata": {},
6
+
"source": [
7
+
"It works on a Windows 10 PC which has an old GPU: GeForce GTX 550 Ti.\n",
8
+
"\n",
9
+
"Simply run the following command (not using docker) in the root of this project:\n",
10
+
"\n",
11
+
"```\n",
12
+
"jupyter notebook\n",
13
+
"```\n",
14
+
"\n",
15
+
"Also run the following in a separate terminal:\n",
16
+
"\n",
17
+
"```\n",
18
+
"tensorboard --logdir=./dl/logs --host=127.0.0.1 --port=8889\n",
19
+
"```"
20
+
]
21
+
},
22
+
{
23
+
"cell_type": "code",
24
+
"execution_count": 1,
25
+
"metadata": {},
26
+
"outputs": [
27
+
{
28
+
"name": "stdout",
29
+
"output_type": "stream",
30
+
"text": [
31
+
"Requirement already satisfied: tensorflow-gpu in c:\\coding\\anaconda3\\lib\\site-packages (1.8.0)\n",
32
+
"Requirement already satisfied: termcolor>=1.1.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (1.1.0)\n",
33
+
"Requirement already satisfied: protobuf>=3.4.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (3.6.0)\n",
34
+
"Requirement already satisfied: six>=1.10.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (1.11.0)\n",
35
+
"Requirement already satisfied: absl-py>=0.1.6 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (0.2.2)\n",
36
+
"Requirement already satisfied: astor>=0.6.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (0.6.2)\n",
37
+
"Requirement already satisfied: wheel>=0.26 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (0.31.1)\n",
38
+
"Requirement already satisfied: tensorboard<1.9.0,>=1.8.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (1.8.0)\n",
39
+
"Requirement already satisfied: grpcio>=1.8.6 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (1.12.1)\n",
40
+
"Requirement already satisfied: numpy>=1.13.3 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (1.14.3)\n",
41
+
"Requirement already satisfied: gast>=0.2.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorflow-gpu) (0.2.0)\n",
42
+
"Requirement already satisfied: setuptools in c:\\coding\\anaconda3\\lib\\site-packages (from protobuf>=3.4.0->tensorflow-gpu) (39.1.0)\n",
43
+
"Requirement already satisfied: markdown>=2.6.8 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard<1.9.0,>=1.8.0->tensorflow-gpu) (2.6.11)\n",
44
+
"Requirement already satisfied: werkzeug>=0.11.10 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard<1.9.0,>=1.8.0->tensorflow-gpu) (0.14.1)\n",
45
+
"Requirement already satisfied: bleach==1.5.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard<1.9.0,>=1.8.0->tensorflow-gpu) (1.5.0)\n",
46
+
"Requirement already satisfied: html5lib==0.9999999 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard<1.9.0,>=1.8.0->tensorflow-gpu) (0.9999999)\n"
47
+
]
48
+
},
49
+
{
50
+
"name": "stderr",
51
+
"output_type": "stream",
52
+
"text": [
53
+
"distributed 1.21.8 requires msgpack, which is not installed.\n"
54
+
]
55
+
}
56
+
],
57
+
"source": [
58
+
"!pip install tensorflow-gpu"
59
+
]
60
+
},
61
+
{
62
+
"cell_type": "code",
63
+
"execution_count": 2,
64
+
"metadata": {},
65
+
"outputs": [
66
+
{
67
+
"name": "stdout",
68
+
"output_type": "stream",
69
+
"text": [
70
+
"env: CUDA_DEVICE_ORDER=PCI_BUS_ID\n",
71
+
"env: CUDA_VISIBLE_DEVICES=0,1\n"
72
+
]
73
+
}
74
+
],
75
+
"source": [
76
+
"%env CUDA_DEVICE_ORDER=PCI_BUS_ID\n",
77
+
"%env CUDA_VISIBLE_DEVICES=0,1"
78
+
]
79
+
},
80
+
{
81
+
"cell_type": "code",
82
+
"execution_count": 3,
83
+
"metadata": {},
84
+
"outputs": [
85
+
{
86
+
"name": "stdout",
87
+
"output_type": "stream",
88
+
"text": [
89
+
"PCI_BUS_ID\n",
90
+
"0,1\n"
91
+
]
92
+
}
93
+
],
94
+
"source": [
95
+
"import os\n",
96
+
"print(os.environ[\"CUDA_DEVICE_ORDER\"])\n",
97
+
"print(os.environ[\"CUDA_VISIBLE_DEVICES\"])"
98
+
]
99
+
},
100
+
{
101
+
"cell_type": "code",
102
+
"execution_count": 4,
103
+
"metadata": {},
104
+
"outputs": [
105
+
{
106
+
"name": "stderr",
107
+
"output_type": "stream",
108
+
"text": [
109
+
"C:\\coding\\Anaconda3\\lib\\site-packages\\h5py\\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
110
+
" from ._conv import register_converters as _register_converters\n"
111
+
]
112
+
}
113
+
],
114
+
"source": [
115
+
"import tensorflow as tf"
116
+
]
117
+
},
118
+
{
119
+
"cell_type": "code",
120
+
"execution_count": 5,
121
+
"metadata": {},
122
+
"outputs": [
123
+
{
124
+
"data": {
125
+
"text/plain": [
126
+
"[name: \"/device:CPU:0\"\n",
127
+
" device_type: \"CPU\"\n",
128
+
" memory_limit: 268435456\n",
129
+
" locality {\n",
130
+
" }\n",
131
+
" incarnation: 8034297251726998311]"
132
+
]
133
+
},
134
+
"execution_count": 5,
135
+
"metadata": {},
136
+
"output_type": "execute_result"
137
+
}
138
+
],
139
+
"source": [
140
+
"from tensorflow.python.client import device_lib\n",
141
+
"\n",
142
+
"device_lib.list_local_devices()"
143
+
]
144
+
},
145
+
{
146
+
"cell_type": "code",
147
+
"execution_count": 6,
148
+
"metadata": {},
149
+
"outputs": [
150
+
{
151
+
"data": {
152
+
"text/plain": [
153
+
"''"
154
+
]
155
+
},
156
+
"execution_count": 6,
157
+
"metadata": {},
158
+
"output_type": "execute_result"
159
+
}
160
+
],
161
+
"source": [
162
+
"tf.test.gpu_device_name()"
163
+
]
164
+
},
165
+
{
166
+
"cell_type": "markdown",
167
+
"metadata": {},
168
+
"source": [
169
+
"From https://mxnet.incubator.apache.org/install/index.html :"
170
+
]
171
+
},
172
+
{
173
+
"cell_type": "code",
174
+
"execution_count": 7,
175
+
"metadata": {},
176
+
"outputs": [
177
+
{
178
+
"name": "stdout",
179
+
"output_type": "stream",
180
+
"text": [
181
+
"nvcc: NVIDIA (R) Cuda compiler driver\n",
182
+
"Copyright (c) 2005-2017 NVIDIA Corporation\n",
183
+
"Built on Fri_Sep__1_21:08:32_Central_Daylight_Time_2017\n",
184
+
"Cuda compilation tools, release 9.0, V9.0.176\n"
185
+
]
186
+
}
187
+
],
188
+
"source": [
189
+
"!nvcc --version"
190
+
]
191
+
},
192
+
{
193
+
"cell_type": "code",
194
+
"execution_count": 8,
195
+
"metadata": {
196
+
"scrolled": true
197
+
},
198
+
"outputs": [
199
+
{
200
+
"name": "stdout",
201
+
"output_type": "stream",
202
+
"text": [
203
+
"Requirement already satisfied: mxnet-cu90 in c:\\coding\\anaconda3\\lib\\site-packages (1.2.0)\n",
204
+
"Requirement already satisfied: requests in c:\\coding\\anaconda3\\lib\\site-packages (from mxnet-cu90) (2.18.4)\n",
205
+
"Requirement already satisfied: numpy in c:\\coding\\anaconda3\\lib\\site-packages (from mxnet-cu90) (1.14.3)\n",
206
+
"Requirement already satisfied: graphviz in c:\\coding\\anaconda3\\lib\\site-packages (from mxnet-cu90) (0.8.3)\n",
207
+
"Requirement already satisfied: chardet<3.1.0,>=3.0.2 in c:\\coding\\anaconda3\\lib\\site-packages (from requests->mxnet-cu90) (3.0.4)\n",
208
+
"Requirement already satisfied: idna<2.7,>=2.5 in c:\\coding\\anaconda3\\lib\\site-packages (from requests->mxnet-cu90) (2.6)\n",
209
+
"Requirement already satisfied: urllib3<1.23,>=1.21.1 in c:\\coding\\anaconda3\\lib\\site-packages (from requests->mxnet-cu90) (1.22)\n",
210
+
"Requirement already satisfied: certifi>=2017.4.17 in c:\\coding\\anaconda3\\lib\\site-packages (from requests->mxnet-cu90) (2018.4.16)\n"
211
+
]
212
+
},
213
+
{
214
+
"name": "stderr",
215
+
"output_type": "stream",
216
+
"text": [
217
+
"distributed 1.21.8 requires msgpack, which is not installed.\n"
218
+
]
219
+
}
220
+
],
221
+
"source": [
222
+
"!pip install mxnet-cu90"
223
+
]
224
+
},
225
+
{
226
+
"cell_type": "code",
227
+
"execution_count": 9,
228
+
"metadata": {},
229
+
"outputs": [
230
+
{
231
+
"name": "stdout",
232
+
"output_type": "stream",
233
+
"text": [
234
+
"timestamp, name, pci.bus_id, driver_version, pstate, pcie.link.gen.max, pcie.link.gen.current, temperature.gpu, utilization.gpu [%], utilization.memory [%], memory.total [MiB], memory.free [MiB], memory.used [MiB]\n",
235
+
"2018/06/29 09:51:58.754, GeForce GTX 550 Ti, 00000000:01:00.0, 385.54, P0, [Not Supported], [Not Supported], 42, [Not Supported], [Not Supported], 1024 MiB, 448 MiB, 576 MiB\n"
236
+
]
237
+
}
238
+
],
239
+
"source": [
240
+
"# From https://stackoverflow.com/questions/49076092/is-there-a-way-to-check-if-mxnet-uses-my-gpu/49079940#49079940\n",
241
+
"# https://developer.download.nvidia.com/compute/DCGM/docs/nvidia-smi-367.38.pdf\n",
242
+
"!\"C:\\Program Files\\NVIDIA Corporation\\NVSMI\\nvidia-smi\" --query-gpu=timestamp,name,pci.bus_id,driver_version,pstate,pcie.link.gen.max,pcie.link.gen.current,temperature.gpu,utilization.gpu,utilization.memory,memory.total,memory.free,memory.used --format=csv"
243
+
]
244
+
},
245
+
{
246
+
"cell_type": "code",
247
+
"execution_count": 10,
248
+
"metadata": {},
249
+
"outputs": [],
250
+
"source": [
251
+
"import mxnet as mx \n",
252
+
"def gpu_device(gpu_number=0):\n",
253
+
" try:\n",
254
+
" _ = mx.nd.array([1, 2, 3], ctx=mx.gpu(gpu_number))\n",
255
+
" except mx.MXNetError:\n",
256
+
" return None\n",
257
+
" return mx.gpu(gpu_number)"
258
+
]
259
+
},
260
+
{
261
+
"cell_type": "code",
262
+
"execution_count": 11,
263
+
"metadata": {},
264
+
"outputs": [
265
+
{
266
+
"data": {
267
+
"text/plain": [
268
+
"gpu(0)"
269
+
]
270
+
},
271
+
"execution_count": 11,
272
+
"metadata": {},
273
+
"output_type": "execute_result"
274
+
}
275
+
],
276
+
"source": [
277
+
"gpu_device()"
278
+
]
279
+
},
280
+
{
281
+
"cell_type": "code",
282
+
"execution_count": 12,
283
+
"metadata": {},
284
+
"outputs": [
285
+
{
286
+
"data": {
287
+
"text/plain": [
288
+
"gpu(0)"
289
+
]
290
+
},
291
+
"execution_count": 12,
292
+
"metadata": {},
293
+
"output_type": "execute_result"
294
+
}
295
+
],
296
+
"source": [
297
+
"mx.gpu(0)"
298
+
]
299
+
},
300
+
{
301
+
"cell_type": "markdown",
302
+
"metadata": {},
303
+
"source": [
304
+
"From https://gluon.mxnet.io/chapter03_deep-neural-networks/mlp-gluon.html :"
305
+
]
306
+
},
307
+
{
308
+
"cell_type": "code",
309
+
"execution_count": 13,
310
+
"metadata": {},
311
+
"outputs": [],
312
+
"source": [
313
+
"from __future__ import print_function\n",
314
+
"import numpy as np\n",
315
+
"import mxnet as mx\n",
316
+
"from mxnet import nd, autograd, gluon"
317
+
]
318
+
},
319
+
{
320
+
"cell_type": "code",
321
+
"execution_count": 14,
322
+
"metadata": {},
323
+
"outputs": [],
324
+
"source": [
325
+
"data_ctx = mx.cpu()\n",
326
+
"model_ctx = mx.cpu()\n",
327
+
"# model_ctx = mx.gpu(0)"
328
+
]
329
+
},
330
+
{
331
+
"cell_type": "code",
332
+
"execution_count": 15,
333
+
"metadata": {},
334
+
"outputs": [],
335
+
"source": [
336
+
"batch_size = 64\n",
337
+
"num_inputs = 784\n",
338
+
"num_outputs = 10\n",
339
+
"num_examples = 60000\n",
340
+
"def transform(data, label):\n",
341
+
" return data.astype(np.float32)/255, label.astype(np.float32)"
342
+
]
343
+
},
344
+
{
345
+
"cell_type": "code",
346
+
"execution_count": 16,
347
+
"metadata": {},
348
+
"outputs": [],
349
+
"source": [
350
+
"train_data = mx.gluon.data.DataLoader(mx.gluon.data.vision.MNIST(train=True, transform=transform),\n",
351
+
" batch_size, shuffle=True)\n",
352
+
"test_data = mx.gluon.data.DataLoader(mx.gluon.data.vision.MNIST(train=False, transform=transform),\n",
353
+
" batch_size, shuffle=False)"
354
+
]
355
+
},
356
+
{
357
+
"cell_type": "code",
358
+
"execution_count": 17,
359
+
"metadata": {},
360
+
"outputs": [],
361
+
"source": [
362
+
"class MLP(gluon.Block):\n",
363
+
" def __init__(self, **kwargs):\n",
364
+
" super(MLP, self).__init__(**kwargs)\n",
365
+
" with self.name_scope():\n",
366
+
" self.dense0 = gluon.nn.Dense(64)\n",
367
+
" self.dense1 = gluon.nn.Dense(64)\n",
368
+
" self.dense2 = gluon.nn.Dense(10)\n",
369
+
"\n",
370
+
" def forward(self, x):\n",
371
+
" x = nd.relu(self.dense0(x))\n",
372
+
" x = nd.relu(self.dense1(x))\n",
373
+
" x = self.dense2(x)\n",
374
+
" return x"
375
+
]
376
+
},
377
+
{
378
+
"cell_type": "code",
379
+
"execution_count": 18,
380
+
"metadata": {},
381
+
"outputs": [],
382
+
"source": [
383
+
"net = MLP()\n",
384
+
"net.collect_params().initialize(mx.init.Normal(sigma=.01), ctx=model_ctx)"
385
+
]
386
+
},
387
+
{
388
+
"cell_type": "code",
389
+
"execution_count": 19,
390
+
"metadata": {},
391
+
"outputs": [
392
+
{
393
+
"data": {
394
+
"text/plain": [
395
+
"\n",
396
+
"[[-5.2642502e-04 -4.8494569e-04 -9.1017238e-05 -1.0700601e-03\n",
397
+
" 9.5340359e-04 1.2931204e-03 -3.8861975e-04 -6.4619188e-04\n",
398
+
" 1.3646495e-04 -1.7153830e-03]]\n",
399
+
"<NDArray 1x10 @cpu(0)>"
400
+
]
401
+
},
402
+
"execution_count": 19,
403
+
"metadata": {},
404
+
"output_type": "execute_result"
405
+
}
406
+
],
407
+
"source": [
408
+
"data = nd.ones((1,784))\n",
409
+
"net(data.as_in_context(model_ctx))"
410
+
]
411
+
},
412
+
{
413
+
"cell_type": "code",
414
+
"execution_count": 20,
415
+
"metadata": {},
416
+
"outputs": [
417
+
{
418
+
"name": "stdout",
419
+
"output_type": "stream",
420
+
"text": [
421
+
"Hidden Representation 1: \n",
422
+
"[[0. 0. 0.0257028 0.41763663 0. 0.\n",
423
+
" 0. 0. 0. 0.03712562 0.16054314 0.35507876\n",
424
+
" 0. 0.12578698 0. 0. 0. 0.30374664\n",
425
+
" 0.292567 0.35357708 0. 0.07809136 0.21969806 0.2177984\n",
426
+
" 0. 0.3457912 0.13206203 0.01624641 0.27534354 0.22952288\n",
427
+
" 0.2202207 0. 0.00258669 0.06395139 0.68015635 0.\n",
428
+
" 0. 0. 0.1652459 0.18695295 0.25243065 0.01728743\n",
429
+
" 0.06471729 0. 0. 0.2552151 0. 0.\n",
430
+
" 0.03300378 0.33107045 0.6453747 0.04547642 0. 0.\n",
431
+
" 0. 0.19542485 0.02424754 0. 0. 0.04300808\n",
432
+
" 0.16542053 0.13203493 0. 0. ]]\n",
433
+
"<NDArray 1x64 @cpu(0)>\n",
434
+
"Hidden Representation 2: \n",
435
+
"[[0.0000000e+00 0.0000000e+00 4.8457514e-03 0.0000000e+00 2.4975553e-02\n",
436
+
" 0.0000000e+00 9.2384806e-03 1.1846514e-02 0.0000000e+00 1.5087268e-02\n",
437
+
" 0.0000000e+00 1.3427198e-02 1.6015759e-02 0.0000000e+00 0.0000000e+00\n",
438
+
" 0.0000000e+00 0.0000000e+00 0.0000000e+00 2.7162414e-02 4.1979598e-05\n",
439
+
" 0.0000000e+00 1.8946800e-02 3.0578913e-03 0.0000000e+00 0.0000000e+00\n",
440
+
" 2.7754948e-02 7.5642066e-04 0.0000000e+00 0.0000000e+00 1.9757828e-02\n",
441
+
" 1.7670706e-02 0.0000000e+00 4.0669916e-03 1.0265570e-02 7.5005908e-03\n",
442
+
" 1.5555882e-02 0.0000000e+00 0.0000000e+00 0.0000000e+00 2.8156085e-02\n",
443
+
" 0.0000000e+00 0.0000000e+00 2.0807199e-02 0.0000000e+00 0.0000000e+00\n",
444
+
" 0.0000000e+00 5.2651879e-04 0.0000000e+00 0.0000000e+00 3.6671013e-02\n",
445
+
" 1.6886523e-02 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00\n",
446
+
" 1.5089142e-02 1.0638590e-02 9.0155248e-03 1.8627236e-02 1.4041221e-02\n",
447
+
" 0.0000000e+00 0.0000000e+00 0.0000000e+00 1.2555162e-02]]\n",
448
+
"<NDArray 1x64 @cpu(0)>\n",
449
+
"Network output: \n",
450
+
"[[-1.1785791e-03 1.9014490e-04 8.1118196e-04 -3.8255830e-04\n",
451
+
" 4.7956721e-04 -1.2719276e-04 3.3852040e-05 -2.3284566e-04\n",
452
+
" 7.1805023e-04 1.1753932e-03]]\n",
453
+
"<NDArray 1x10 @cpu(0)>\n"
454
+
]
455
+
},
456
+
{
457
+
"data": {
458
+
"text/plain": [
459
+
"\n",
460
+
"[[-1.1785791e-03 1.9014490e-04 8.1118196e-04 -3.8255830e-04\n",
461
+
" 4.7956721e-04 -1.2719276e-04 3.3852040e-05 -2.3284566e-04\n",
462
+
" 7.1805023e-04 1.1753932e-03]]\n",
463
+
"<NDArray 1x10 @cpu(0)>"
464
+
]
465
+
},
466
+
"execution_count": 20,
467
+
"metadata": {},
468
+
"output_type": "execute_result"
469
+
}
470
+
],
471
+
"source": [
472
+
"class MLP(gluon.Block):\n",
473
+
" def __init__(self, **kwargs):\n",
474
+
" super(MLP, self).__init__(**kwargs)\n",
475
+
" with self.name_scope():\n",
476
+
" self.dense0 = gluon.nn.Dense(64, activation=\"relu\")\n",
477
+
" self.dense1 = gluon.nn.Dense(64, activation=\"relu\")\n",
478
+
" self.dense2 = gluon.nn.Dense(10)\n",
479
+
"\n",
480
+
" def forward(self, x):\n",
481
+
" x = self.dense0(x)\n",
482
+
" print(\"Hidden Representation 1: %s\" % x)\n",
483
+
" x = self.dense1(x)\n",
484
+
" print(\"Hidden Representation 2: %s\" % x)\n",
485
+
" x = self.dense2(x)\n",
486
+
" print(\"Network output: %s\" % x)\n",
487
+
" return x\n",
488
+
"\n",
489
+
"net = MLP()\n",
490
+
"net.collect_params().initialize(mx.init.Normal(sigma=.01), ctx=model_ctx)\n",
491
+
"net(data.as_in_context(model_ctx))"
492
+
]
493
+
},
494
+
{
495
+
"cell_type": "code",
496
+
"execution_count": 33,
497
+
"metadata": {},
498
+
"outputs": [],
499
+
"source": [
500
+
"num_hidden = 64\n",
501
+
"net = gluon.nn.HybridSequential()\n",
502
+
"with net.name_scope():\n",
503
+
" net.add(gluon.nn.Dense(num_hidden, activation=\"relu\"))\n",
504
+
" net.add(gluon.nn.Dense(num_hidden, activation=\"relu\"))\n",
505
+
" net.add(gluon.nn.Dense(num_outputs))\n",
506
+
"net.hybridize()"
507
+
]
508
+
},
509
+
{
510
+
"cell_type": "code",
511
+
"execution_count": 34,
512
+
"metadata": {},
513
+
"outputs": [],
514
+
"source": [
515
+
"net.collect_params().initialize(mx.init.Normal(sigma=.1), ctx=model_ctx)"
516
+
]
517
+
},
518
+
{
519
+
"cell_type": "code",
520
+
"execution_count": 35,
521
+
"metadata": {},
522
+
"outputs": [],
523
+
"source": [
524
+
"softmax_cross_entropy = gluon.loss.SoftmaxCrossEntropyLoss()"
525
+
]
526
+
},
527
+
{
528
+
"cell_type": "code",
529
+
"execution_count": 36,
530
+
"metadata": {},
531
+
"outputs": [],
532
+
"source": [
533
+
"trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': .01})"
534
+
]
535
+
},
536
+
{
537
+
"cell_type": "code",
538
+
"execution_count": 37,
539
+
"metadata": {},
540
+
"outputs": [],
541
+
"source": [
542
+
"def evaluate_accuracy(data_iterator, net):\n",
543
+
" acc = mx.metric.Accuracy()\n",
544
+
" for i, (data, label) in enumerate(data_iterator):\n",
545
+
" data = data.as_in_context(model_ctx).reshape((-1, 784))\n",
546
+
" label = label.as_in_context(model_ctx)\n",
547
+
" output = net(data)\n",
548
+
" predictions = nd.argmax(output, axis=1)\n",
549
+
" acc.update(preds=predictions, labels=label)\n",
550
+
" return acc.get()[1]"
551
+
]
552
+
},
553
+
{
554
+
"cell_type": "code",
555
+
"execution_count": 38,
556
+
"metadata": {},
557
+
"outputs": [
558
+
{
559
+
"name": "stdout",
560
+
"output_type": "stream",
561
+
"text": [
562
+
"Requirement already satisfied: mxboard in c:\\coding\\anaconda3\\lib\\site-packages (0.1.0)\n",
563
+
"Requirement already satisfied: protobuf>=3.0.0 in c:\\coding\\anaconda3\\lib\\site-packages (from mxboard) (3.6.0)\n",
564
+
"Requirement already satisfied: Pillow in c:\\coding\\anaconda3\\lib\\site-packages (from mxboard) (5.1.0)\n",
565
+
"Requirement already satisfied: six in c:\\coding\\anaconda3\\lib\\site-packages (from mxboard) (1.11.0)\n",
566
+
"Requirement already satisfied: numpy in c:\\coding\\anaconda3\\lib\\site-packages (from mxboard) (1.14.3)\n",
567
+
"Requirement already satisfied: setuptools in c:\\coding\\anaconda3\\lib\\site-packages (from protobuf>=3.0.0->mxboard) (39.1.0)\n"
568
+
]
569
+
},
570
+
{
571
+
"name": "stderr",
572
+
"output_type": "stream",
573
+
"text": [
574
+
"distributed 1.21.8 requires msgpack, which is not installed.\n"
575
+
]
576
+
}
577
+
],
578
+
"source": [
579
+
"!pip install mxboard"
580
+
]
581
+
},
582
+
{
583
+
"cell_type": "code",
584
+
"execution_count": 39,
585
+
"metadata": {},
586
+
"outputs": [
587
+
{
588
+
"name": "stdout",
589
+
"output_type": "stream",
590
+
"text": [
591
+
"Requirement already satisfied: tensorboard in c:\\coding\\anaconda3\\lib\\site-packages (1.8.0)\n",
592
+
"Requirement already satisfied: html5lib==0.9999999 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard) (0.9999999)\n",
593
+
"Requirement already satisfied: bleach==1.5.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard) (1.5.0)\n",
594
+
"Requirement already satisfied: protobuf>=3.4.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard) (3.6.0)\n",
595
+
"Requirement already satisfied: werkzeug>=0.11.10 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard) (0.14.1)\n",
596
+
"Requirement already satisfied: six>=1.10.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard) (1.11.0)\n",
597
+
"Requirement already satisfied: wheel>=0.26; python_version >= \"3\" in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard) (0.31.1)\n",
598
+
"Requirement already satisfied: markdown>=2.6.8 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard) (2.6.11)\n",
599
+
"Requirement already satisfied: numpy>=1.12.0 in c:\\coding\\anaconda3\\lib\\site-packages (from tensorboard) (1.14.3)\n",
600
+
"Requirement already satisfied: setuptools in c:\\coding\\anaconda3\\lib\\site-packages (from protobuf>=3.4.0->tensorboard) (39.1.0)\n"
601
+
]
602
+
},
603
+
{
604
+
"name": "stderr",
605
+
"output_type": "stream",
606
+
"text": [
607
+
"distributed 1.21.8 requires msgpack, which is not installed.\n"
608
+
]
609
+
}
610
+
],
611
+
"source": [
612
+
"!pip install tensorboard"
613
+
]
614
+
},
615
+
{
616
+
"cell_type": "code",
617
+
"execution_count": 40,
618
+
"metadata": {},
619
+
"outputs": [],
620
+
"source": [
621
+
"from mxboard import SummaryWriter\n",
622
+
"sw = SummaryWriter(logdir='logs', flush_secs=5)"
623
+
]
624
+
},
625
+
{
626
+
"cell_type": "code",
627
+
"execution_count": 42,
628
+
"metadata": {},
629
+
"outputs": [
630
+
{
631
+
"name": "stdout",
632
+
"output_type": "stream",
633
+
"text": [
634
+
"Epoch 0. Loss: 1.2356886094411215, Train_acc 0.8395166666666667, Test_acc 0.8474\n",
635
+
"Epoch 1. Loss: 0.46565542084376016, Train_acc 0.88465, Test_acc 0.8912\n",
636
+
"Epoch 2. Loss: 0.3715192502895991, Train_acc 0.901, Test_acc 0.9029\n",
637
+
"Epoch 3. Loss: 0.32939287207126616, Train_acc 0.91, Test_acc 0.9112\n",
638
+
"Epoch 4. Loss: 0.30172612600326537, Train_acc 0.91625, Test_acc 0.9184\n",
639
+
"Epoch 5. Loss: 0.2804558823386828, Train_acc 0.9217333333333333, Test_acc 0.921\n",
640
+
"Epoch 6. Loss: 0.2626380964756012, Train_acc 0.9262666666666667, Test_acc 0.9252\n",
641
+
"Epoch 7. Loss: 0.24708882774909338, Train_acc 0.9300333333333334, Test_acc 0.9296\n",
642
+
"Epoch 8. Loss: 0.23393845278819403, Train_acc 0.9341833333333334, Test_acc 0.9332\n",
643
+
"Epoch 9. Loss: 0.22262413431803385, Train_acc 0.93545, Test_acc 0.9352\n"
644
+
]
645
+
}
646
+
],
647
+
"source": [
648
+
"epochs = 10\n",
649
+
"smoothing_constant = .01\n",
650
+
"\n",
651
+
"# collect parameter names for logging the gradients of parameters in each epoch\n",
652
+
"params = net.collect_params()\n",
653
+
"param_names = params.keys()\n",
654
+
"global_step = 0\n",
655
+
"\n",
656
+
"for e in range(epochs):\n",
657
+
" cumulative_loss = 0\n",
658
+
" for i, (data, label) in enumerate(train_data):\n",
659
+
" data = data.as_in_context(model_ctx).reshape((-1, 784))\n",
660
+
" label = label.as_in_context(model_ctx)\n",
661
+
" with autograd.record():\n",
662
+
" output = net(data)\n",
663
+
" loss = softmax_cross_entropy(output, label)\n",
664
+
" \n",
665
+
" sw.add_scalar(tag='cross_entropy', value=loss.mean().asscalar(), global_step=global_step)\n",
666
+
" if i == 0:\n",
667
+
" sw.add_image('minist_first_minibatch', data.reshape((batch_size, 1, 28, 28)), e)\n",
668
+
" if e == 0:\n",
669
+
" sw.add_graph(net)\n",
670
+
" grads = [i.grad() for i in net.collect_params().values()]\n",
671
+
" for i, name in enumerate(param_names):\n",
672
+
" sw.add_histogram(tag=name, values=grads[i], global_step=e, bins=1000)\n",
673
+
"\n",
674
+
" global_step += 1\n",
675
+
" loss.backward()\n",
676
+
" trainer.step(data.shape[0])\n",
677
+
" cumulative_loss += nd.sum(loss).asscalar()\n",
678
+
"\n",
679
+
" \n",
680
+
" \n",
681
+
" test_accuracy = evaluate_accuracy(test_data, net)\n",
682
+
" train_accuracy = evaluate_accuracy(train_data, net)\n",
683
+
" sw.add_scalar(tag='accuracy_curves', value=('train_acc', train_accuracy), global_step=e)\n",
684
+
" sw.add_scalar(tag='accuracy_curves', value=('valid_acc', test_accuracy), global_step=e)\n",
685
+
" print(\"Epoch %s. Loss: %s, Train_acc %s, Test_acc %s\" %\n",
686
+
" (e, cumulative_loss/num_examples, train_accuracy, test_accuracy))"
687
+
]
688
+
},
689
+
{
690
+
"cell_type": "markdown",
691
+
"metadata": {},
692
+
"source": [
693
+
"\n",
694
+
"\n",
695
+
"\n",
696
+
"\n",
697
+
"\n",
698
+
""
699
+
]
700
+
},
701
+
{
702
+
"cell_type": "code",
703
+
"execution_count": null,
704
+
"metadata": {},
705
+
"outputs": [],
706
+
"source": []
707
+
}
708
+
],
709
+
"metadata": {
710
+
"kernelspec": {
711
+
"display_name": "Python 3",
712
+
"language": "python",
713
+
"name": "python3"
714
+
},
715
+
"language_info": {
716
+
"codemirror_mode": {
717
+
"name": "ipython",
718
+
"version": 3
719
+
},
720
+
"file_extension": ".py",
721
+
"mimetype": "text/x-python",
722
+
"name": "python",
723
+
"nbconvert_exporter": "python",
724
+
"pygments_lexer": "ipython3",
725
+
"version": "3.6.5"
726
+
}
727
+
},
728
+
"nbformat": 4,
729
+
"nbformat_minor": 2
730
+
}
dl/images/mx_01_acc.png
dl/images/mx_01_acc.png
This is a binary file and will not be displayed.
dl/images/mx_02_loss.png
dl/images/mx_02_loss.png
This is a binary file and will not be displayed.
dl/images/mx_03_img.png
dl/images/mx_03_img.png
This is a binary file and will not be displayed.
dl/images/mx_04_net.png
dl/images/mx_04_net.png
This is a binary file and will not be displayed.
dl/images/mx_05_dist.png
dl/images/mx_05_dist.png
This is a binary file and will not be displayed.
dl/images/mx_06_hist.png
dl/images/mx_06_hist.png
This is a binary file and will not be displayed.