forked from fastai/fastai
/
train.html
452 lines (322 loc) · 18.2 KB
/
train.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
---
title: train
keywords: fastai
sidebar: home_sidebar
summary: "Extensions to Learner that easily implement Callback"
---
<div class="container" id="notebook-container">
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<h1 id="Additional-training-functions">Additional training functions<a class="anchor-link" href="#Additional-training-functions">¶</a></h1>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p><a href="/train.html#train"><code>train</code></a> provides a number of extension methods that are added to <a href="/basic_train.html#Learner"><code>Learner</code></a> (see below for a list and details), along with three simple callbacks:</p>
<ul>
<li><a href="/train.html#ShowGraph"><code>ShowGraph</code></a></li>
<li><a href="/train.html#GradientClipping"><code>GradientClipping</code></a></li>
<li><a href="/train.html#BnFreeze"><code>BnFreeze</code></a></li>
</ul>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<h2 id="Learner-extension-methods"><a href="/basic_train.html#Learner"><code>Learner</code></a> extension methods<a class="anchor-link" href="#Learner-extension-methods">¶</a></h2>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>These methods are automatically added to all <a href="/basic_train.html#Learner"><code>Learner</code></a> objects created after importing this module. They provide convenient access to a number of callbacks, without requiring them to be manually created.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h4 id="fit_one_cycle"><code>fit_one_cycle</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L14" class="source_link">[source]</a></h4><blockquote><p><code>fit_one_cycle</code>(<code>learn</code>:<a href="/basic_train.html#Learner"><code>Learner</code></a>, <code>cyc_len</code>:<code>int</code>, <code>max_lr</code>:<code>Union</code>[<code>float</code>, <code>Collection</code>[<code>float</code>], <code>slice</code>]=<code>slice(None, 0.003, None)</code>, <code>moms</code>:<code>Point</code>=<code>(0.95, 0.85)</code>, <code>div_factor</code>:<code>float</code>=<code>25.0</code>, <code>pct_start</code>:<code>float</code>=<code>0.3</code>, <code>wd</code>:<code>float</code>=<code>None</code>, <code>callbacks</code>:<code>Optional</code>[<code>Collection</code>[<a href="/callback.html#Callback"><code>Callback</code></a>]]=<code>None</code>, <code>kwargs</code>)</p>
</blockquote>
<p>Fit a model following the 1cycle policy.</p>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>Fit a model with 1cycle training. See <a href="/callbacks.one_cycle.html#OneCycleScheduler"><code>OneCycleScheduler</code></a> for details.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h4 id="lr_find"><code>lr_find</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L24" class="source_link">[source]</a></h4><blockquote><p><code>lr_find</code>(<code>learn</code>:<a href="/basic_train.html#Learner"><code>Learner</code></a>, <code>start_lr</code>:<code>Floats</code>=<code>1e-07</code>, <code>end_lr</code>:<code>Floats</code>=<code>10</code>, <code>num_it</code>:<code>int</code>=<code>100</code>, <code>stop_div</code>:<code>bool</code>=<code>True</code>, <code>kwargs</code>:<code>Any</code>)</p>
</blockquote>
<p>Explore lr from <code>start_lr</code> to <code>end_lr</code> over <code>num_it</code> iterations in <code>learn</code>. If <code>stop_div</code>, stops when loss explodes.</p>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>See <a href="/callbacks.lr_finder.html#LRFinder"><code>LRFinder</code></a> for details.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h4 id="to_fp16"><code>to_fp16</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L32" class="source_link">[source]</a></h4><blockquote><p><code>to_fp16</code>(<code>learn</code>:<a href="/basic_train.html#Learner"><code>Learner</code></a>, <code>loss_scale</code>:<code>float</code>=<code>512.0</code>, <code>flat_master</code>:<code>bool</code>=<code>False</code>) → <a href="/basic_train.html#Learner"><code>Learner</code></a></p>
</blockquote>
<p>Transform <code>learn</code> in FP16 precision.</p>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>See <a href="/callbacks.fp16.html#MixedPrecision"><code>MixedPrecision</code></a> for details.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h4 id="mixup"><code>mixup</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L39" class="source_link">[source]</a></h4><blockquote><p><code>mixup</code>(<code>learn</code>:<a href="/basic_train.html#Learner"><code>Learner</code></a>, <code>alpha</code>:<code>float</code>=<code>0.4</code>, <code>stack_x</code>:<code>bool</code>=<code>False</code>, <code>stack_y</code>:<code>bool</code>=<code>True</code>) → <a href="/basic_train.html#Learner"><code>Learner</code></a></p>
</blockquote>
<p>Add mixup <a href="https://arxiv.org/abs/1710.09412">https://arxiv.org/abs/1710.09412</a> to <code>learn</code>.</p>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>See <a href="/callbacks.mixup.html#MixUpCallback"><code>MixUpCallback</code></a> for more details.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>A last extension method comes from the module tta.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h4 id="TTA"><code>TTA</code><a href="https://github.com/fastai/fastai/blob/master/fastai/vision/tta.py#L32" class="source_link">[source]</a></h4><blockquote><p><code>TTA</code>(<code>learn</code>:<a href="/basic_train.html#Learner"><code>Learner</code></a>, <code>beta</code>:<code>float</code>=<code>0.4</code>, <code>scale</code>:<code>float</code>=<code>1.35</code>, <code>ds_type</code>:<a href="/basic_data.html#DatasetType"><code>DatasetType</code></a>=<code><DatasetType.Valid: 2></code>, <code>with_loss</code>:<code>bool</code>=<code>False</code>) → <code>Tensors</code></p>
</blockquote>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>Applies Test Time Augmentation to <code>learn</code> on the dataset <code>ds_type</code>. We take the average of our regular predictions (with a weight <code>beta</code>) with the average of predictions obtained thourh augmented versions of the training set (with a weight <code>1-beta</code>). The transforms decided for the training set are applied with a few changes <code>scale</code> controls the scale for zoom (which isn't random), the cropping isn't random but we make sure to get the four corners of the image. Flipping isn't random but applied once on each of those corner images (so that makes 8 augmented versions total).</p>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>We'll show examples below using our MNIST sample.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="inner_cell">
<div class="input_area">
<div class=" highlight hl-ipython3"><pre><span></span><span class="n">path</span> <span class="o">=</span> <span class="n">untar_data</span><span class="p">(</span><span class="n">URLs</span><span class="o">.</span><span class="n">MNIST_SAMPLE</span><span class="p">)</span>
<span class="n">data</span> <span class="o">=</span> <span class="n">ImageDataBunch</span><span class="o">.</span><span class="n">from_folder</span><span class="p">(</span><span class="n">path</span><span class="p">)</span>
</pre></div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h2 id="ShowGraph"><code>class</code> <code>ShowGraph</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L50" class="source_link">[source]</a></h2><blockquote><p><code>ShowGraph</code>(<code>learn</code>:<a href="/basic_train.html#Learner"><code>Learner</code></a>) :: <a href="/basic_train.html#LearnerCallback"><code>LearnerCallback</code></a></p>
</blockquote>
<p>Update a graph of learner stats and metrics after each epoch.</p>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<div class="highlight"><pre><span></span><span class="n">learn</span> <span class="o">=</span> <span class="n">create_cnn</span><span class="p">(</span><span class="n">data</span><span class="p">,</span> <span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">,</span> <span class="n">metrics</span><span class="o">=</span><span class="n">accuracy</span><span class="p">,</span> <span class="n">callback_fns</span><span class="o">=</span><span class="n">ShowGraph</span><span class="p">)</span>
<span class="n">learn</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span>
</pre></div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p><img src="imgs/train_graph.gif" alt="Training graph"></p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h4 id="ShowGraph.on_epoch_end"><code>on_epoch_end</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L52" class="source_link">[source]</a></h4><blockquote><p><code>on_epoch_end</code>(<code>n_epochs</code>:<code>int</code>, <code>last_metrics</code>:<code>MetricsList</code>, <code>kwargs</code>) → <code>bool</code></p>
</blockquote>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>If we have <code>last_metrics</code>, plot them in <code>self.pbar</code>. Set the size of the graph with <code>n_epochs</code>.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h2 id="GradientClipping"><code>class</code> <code>GradientClipping</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L70" class="source_link">[source]</a></h2><blockquote><p><code>GradientClipping</code>(<code>learn</code>:<a href="/basic_train.html#Learner"><code>Learner</code></a>, <code>clip</code>:<code>float</code>) :: <a href="/basic_train.html#LearnerCallback"><code>LearnerCallback</code></a></p>
</blockquote>
<p>To do gradient clipping during training.</p>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>Clips gradient at a maximum absolute value of <code>clip</code> during training. For instance:</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="inner_cell">
<div class="input_area">
<div class=" highlight hl-ipython3"><pre><span></span><span class="n">learn</span> <span class="o">=</span> <span class="n">create_cnn</span><span class="p">(</span><span class="n">data</span><span class="p">,</span> <span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">,</span> <span class="n">metrics</span><span class="o">=</span><span class="n">accuracy</span><span class="p">,</span>
<span class="n">callback_fns</span><span class="o">=</span><span class="n">partial</span><span class="p">(</span><span class="n">GradientClipping</span><span class="p">,</span> <span class="n">clip</span><span class="o">=</span><span class="mf">0.1</span><span class="p">))</span>
<span class="n">learn</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
</pre></div>
</div>
</div>
</div>
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_subarea output_stream output_stdout output_text">
<pre>Total time: 00:11
epoch train loss valid loss accuracy
0 0.086958 0.038721 0.989696 (00:11)
</pre>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h4 id="GradientClipping.on_backward_end"><code>on_backward_end</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L74" class="source_link">[source]</a></h4><blockquote><p><code>on_backward_end</code>(<code>kwargs</code>)</p>
</blockquote>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>Clip the gradients after they are computed but before the optimizer step.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h2 id="BnFreeze"><code>class</code> <code>BnFreeze</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L63" class="source_link">[source]</a></h2><blockquote><p><code>BnFreeze</code>(<code>learn</code>:<a href="/basic_train.html#Learner"><code>Learner</code></a>) :: <a href="/basic_train.html#LearnerCallback"><code>LearnerCallback</code></a></p>
</blockquote>
<p>Freeze moving average statistics in all non-trainable batchnorm layers.</p>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>For batchnorm layers where <code>requires_grad==False</code>, you generally don't want to update their moving average statistics, in order to avoid the model's statistics getting out of sync with its pre-trained weights. You can add this callback to automate this freezing of statistics (internally, it calls <code>eval</code> on these layers).</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="inner_cell">
<div class="input_area">
<div class=" highlight hl-ipython3"><pre><span></span><span class="n">learn</span> <span class="o">=</span> <span class="n">create_cnn</span><span class="p">(</span><span class="n">data</span><span class="p">,</span> <span class="n">models</span><span class="o">.</span><span class="n">resnet18</span><span class="p">,</span> <span class="n">metrics</span><span class="o">=</span><span class="n">accuracy</span><span class="p">,</span> <span class="n">callback_fns</span><span class="o">=</span><span class="n">BnFreeze</span><span class="p">)</span>
<span class="n">learn</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
</pre></div>
</div>
</div>
</div>
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_subarea output_stream output_stdout output_text">
<pre>Total time: 00:07
epoch train loss valid loss accuracy
0 0.079278 0.041832 0.985280 (00:07)
</pre>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="output_wrapper">
<div class="output">
<div class="output_area">
<div class="output_markdown rendered_html output_subarea ">
<h4 id="BnFreeze.on_epoch_begin"><code>on_epoch_begin</code><a href="https://github.com/fastai/fastai/blob/master/fastai/train.py#L65" class="source_link">[source]</a></h4><blockquote><p><code>on_epoch_begin</code>(<code>kwargs</code>:<code>Any</code>)</p>
</blockquote>
</div>
</div>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>Set back the batchnorm layers on <code>eval</code> mode after the model has been set to <a href="/train.html#train"><code>train</code></a>.</p>
</div>
</div>
</div>
</div>