These are chat archives for thunder-project/thunder

Apr 2015
Jeremy Freeman
Apr 14 2015 00:25
Awesome, I think you've nailed it @ahwillia
The padding can be derived from existing fields, because there's a imgSlices and a padImgSlices
essentially, the padding is the difference between the two
for example, on a dummy test I just made with 10,10,1 blocks and a padding of 10, i get
imgSlices: (slice(0, 20, None), slice(10, 20, 1), slice(0, 10, 1), slice(0, 1, 1))
padImgSlices: (slice(0, 20, None), slice(0, 30, 1), slice(0, 20, 1), slice(0, 2, 1))
the first is time so they are identical, but for the first spatial dimension, you'd get a padding of 10-0=10 and 30-20=10, etc.
so maybe add a @property to PaddedBlockGroupingKey that computes and returns the padding from those existing attributes?
Jeremy Freeman
Apr 14 2015 00:32
that prevents storing the essentially same information in two distinct attributes, which has the potential for confusion if only one is changed
Alex Williams
Apr 14 2015 14:37
This works... but under what conditions will there be at least two entries?
def padding(self):
  return asarray((self.imgSlices[1].start - self.padImgSlices[1].start, self.padImgSlices[1].stop - self.imgSlices[1].stop))
Jason Wittenbach
Apr 14 2015 15:35
@freeman-lab Just pushed the first functional prototype of a new layout for the regression code to my fork if you want to take a look and let me know what you think:
Happy to hear comments from anyone else who might be interested as well
Jeremy Freeman
Apr 14 2015 17:52
@jwittenbach awesome! I'll take a look.