Improve speed of NaturalIdPartitioner #3276
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
adam-narozniak
requested review from
jafermarq,
tanertopal and
danieljanes
as code owners
jafermarq
reviewed
Apr 25, 2024
jafermarq
reviewed
Apr 25, 2024
Comment on lines
63
to
64
| natural_ids = np.array(self.dataset[self._partition_by]) | ||
| unique_natural_ids = self.dataset.unique(self._partition_by) |
There was a problem hiding this comment.
Could we do this without having to load the whole dataset into memory?
There was a problem hiding this comment.
To make it work quicker we need to load the column specified by self._partition_by to the memory.
…er' into fds-improve-natural-id-partitioner
jafermarq
reviewed
Apr 30, 2024
Co-authored-by: Javier <[email protected]>
jafermarq
reviewed
May 6, 2024
jafermarq
approved these changes
May 6, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Issue
When adding and testing the
femnistdataset onNaturalIdPartitioner, I discovered that partitioning takes a very long time.Description
The
femnistdataset consists of over 800,000 samples corresponding to over 3,500 unique writers.The current implementation filters (using
Dataset.filter) the dataset for each unique writer (when there's a request for a particular partition, it filters that for a single unique value). It took about 1 minute to filter the data for a single unique writer_id (it takes 3600 minutes to do that for all = 60 hours). Also, the filtering takes all columns instead of a single column.Proposal
Iterate a single time over the whole column specified by
partition_byto create a mapping.Time
Time old = time it takes to load 100 partitions (using slight optimization of the old way = still using filter)
time(load 100) ~ 100 * time(load 1)and generallytime(load n) = n*time(load 1)Time new = the same as time old, but here the computation happens once, so
time(load 100) ~ time(load 1) ~ time(load all_partitions)Table 1: Table reporting time in seconds to load 100 partitions using an old (with slight modification but still using a filter) and a new implementation. It uses a few datasets that vary in terms of the total number of samples and the number of unique clients.
Changelog entry