Adding a reset_gpu method to the BigWigCollection and Dataset objects… #3
Conversation
…. This can be handy when an existing dataset or collection object should be brought to a new gpu device.
bigwig_loader/collection.py
Outdated
| def _get_decoder(self) -> Decoder: | ||
| if self._decoder: | ||
| return self._decoder | ||
| self._decoder = Decoder( | ||
| max_rows_per_chunk=self.max_rows_per_chunk, | ||
| max_uncompressed_chunk_size=self.max_rows_per_chunk * 12 + 24, | ||
| chromosome_offsets=self.local_chrom_ids_to_offset_matrix, | ||
| ) | ||
| return self._decoder |
There was a problem hiding this comment.
You could use a cached_property here to simplify this a bit.
You'd just need to switch your code in reset_gpu to delete from __dict__
There was a problem hiding this comment.
So, I am mostly using this pattern because mypy tends to complain when I set (and type) self._decoder in init and then have a property:
@property
def decoder(self):
...I can be wrong. Does cached_property fix exactly that problem?
There was a problem hiding this comment.
yep.
You get rid of self._decoder and you just have:
@cached_property
def decoder(self) -> Decoder:
return Decoder(...)Clearing the cache looks like this:
def clear(self):
if "decoder" in self.__dict__:
del self.__dict__["decoder"]It's useful to do it like that if all you want is late initialization of self.decoder
| @@ -75,6 +73,27 @@ def run_indexing(self) -> None: | |||
| def __len__(self) -> int: | |||
| return len(self.bigwigs) | |||
|
|
|||
| def reset_gpu(self) -> None: | |||
There was a problem hiding this comment.
Is this ever called on the collection level?
It seems you have to repeat the method to have to delegate from the top levels down.
There was a problem hiding this comment.
Yes. Because a BigWigDataset object can take a BigWigCollection object as one of the init args. This can be handy when you want to create train, val and test datasets, all based on the same collection of bigwig files, but with different "regions of interest" from which the samples are taken:
…. This can be handy when an existing dataset or collection object should be brought to a new gpu device.