zhreshold commented on a change in pull request #16175: [Dataset] add shard API
URL: https://github.com/apache/incubator-mxnet/pull/16175#discussion_r324829139
##########
File path: python/mxnet/gluon/data/dataset.py
##########
@@ -64,6 +64,37 @@ def filter(self, fn):
from . import FilterSampler
return _SampledDataset(self, FilterSampler(fn, self))
+ def shard(self, num_shards, index):
+ """Returns a new dataset includes only 1/num_shards of this dataset.
+
+ For distributed training, be sure to shard before you randomize the
dataset
+ (such as shuffle), if you want each worker to reach a unique subset.
+
+ Parameters
+ ----------
+ num_shards : int
+ A integer representing the number of data shards.
+ index : int
+ A integer representing the index of the current shard.
+
+ Returns
+ -------
+ Dataset
+ The result dataset.
+ """
+ assert index < num_shards, 'Shard index of out bound: %d out of
%d'%(index, num_shards)
+ assert num_shards > 0, 'Number of shards must be greater than 0'
+ assert index >= 0, 'Index must be non-negative'
+ length = len(self)
+ shard_len = length // num_shards
+ rest = length % num_shards
+ # Compute the start index for this partition
+ start = shard_len * index + min(index, rest)
Review comment:
don't quite get it, why start and end are computed like this?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
