Mercurial > pylearn
comparison dataset.py @ 331:52aa031e1fe3
IMPORTANT: minibatches now returns minibatch_nowrap with a minimum of assert before. Should implement the good behavior, e.g. returning only complete batches and let the user figure out what he wants.
| author | Thierry Bertin-Mahieux <bertinmt@iro.umontreal.ca> |
|---|---|
| date | Mon, 16 Jun 2008 16:38:03 -0400 |
| parents | 20e08c52c98c |
| children | dada08a6adb8 |
comparison
equal
deleted
inserted
replaced
| 330:20e08c52c98c | 331:52aa031e1fe3 |
|---|---|
| 366 | 366 |
| 367 - minibatch_size (integer, default 1) | 367 - minibatch_size (integer, default 1) |
| 368 On every iteration, the variables i1, i2, i3 will have | 368 On every iteration, the variables i1, i2, i3 will have |
| 369 exactly minibatch_size elements. e.g. len(i1) == minibatch_size | 369 exactly minibatch_size elements. e.g. len(i1) == minibatch_size |
| 370 | 370 |
| 371 @DEPRECATED n_batches : not used anywhere | |
| 371 - n_batches (integer, default None) | 372 - n_batches (integer, default None) |
| 372 The iterator will loop exactly this many times, and then stop. If None, | 373 The iterator will loop exactly this many times, and then stop. If None, |
| 373 the derived class can choose a default. If (-1), then the returned | 374 the derived class can choose a default. If (-1), then the returned |
| 374 iterator should support looping indefinitely. | 375 iterator should support looping indefinitely. |
| 375 | 376 |
| 377 The iterator will start at example 'offset' in the dataset, rather than the default. | 378 The iterator will start at example 'offset' in the dataset, rather than the default. |
| 378 | 379 |
| 379 Note: A list-like container is something like a tuple, list, numpy.ndarray or | 380 Note: A list-like container is something like a tuple, list, numpy.ndarray or |
| 380 any other object that supports integer indexing and slicing. | 381 any other object that supports integer indexing and slicing. |
| 381 | 382 |
| 382 """ | 383 @ATTENTION: now minibatches returns minibatches_nowrap, which is supposed to return complete |
| 383 return DataSet.MinibatchWrapAroundIterator(self,fieldnames,minibatch_size,n_batches,offset) | 384 batches only, raise StopIteration |
| 385 | |
| 386 """ | |
| 387 #return DataSet.MinibatchWrapAroundIterator(self,fieldnames,minibatch_size,n_batches,offset)\ | |
| 388 assert offset >= 0 | |
| 389 assert offset < len(self) | |
| 390 assert offset + minibatch_size < len(self) | |
| 391 return minibatch_nowrap(fieldnames,minibatch_size,n_batches,offset) | |
| 384 | 392 |
| 385 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): | 393 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
| 386 """ | 394 """ |
| 387 This is the minibatches iterator generator that sub-classes must define. | 395 This is the minibatches iterator generator that sub-classes must define. |
| 388 It does not need to worry about wrapping around multiple times across the dataset, | 396 It does not need to worry about wrapping around multiple times across the dataset, |