[
https://issues.apache.org/jira/browse/SYSTEMML-951?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15567366#comment-15567366
]
Matthias Boehm commented on SYSTEMML-951:
-----------------------------------------
1) that's again a good catch as this function inherits from a scala function
not spark.java function (which are all serializable). Could you please push
your change into the repo?
2) I would recommend to convert the data frame to a matrix. Y can be fed in
separately, as it is in comparison so small, that we anyway can do random
access in memory. If you intend to run different algorithms or parameters over
the same data, then yes it makes sense to convert the data frame once to a
matrix because this involves shuffle. However, note that currently the
partitioning gets lots when written/read to/from HDFS - so if ran multiple
times through mlcontext and cached externally it's fine; otherwise the explicit
conversion does not help (and can even hurt). One idea to overcome this hdfs
limitation would be to automatically investigate binary input matrices (via a
shuffle free scan), and explicitly set the partitioner if already hash
partitioned (but not marked as such).
> Efficient spark right indexing via lookup
> -----------------------------------------
>
> Key: SYSTEMML-951
> URL: https://issues.apache.org/jira/browse/SYSTEMML-951
> Project: SystemML
> Issue Type: Task
> Components: Runtime
> Reporter: Matthias Boehm
> Assignee: Matthias Boehm
> Attachments: mnist_softmax_v1.dml, mnist_softmax_v2.dml
>
>
> So far all versions of spark right indexing instructions require a full scan
> over the data set. In case of existing partitioning (which anyway happens for
> any external format - binary block conversion) such a full scan is
> unnecessary if we're only interested in a small subset of the data. This task
> adds an efficient right indexing operation via 'rdd lookups' which access at
> most <num_lookup> partitions given existing hash partitioning.
> cc [~mwdus...@us.ibm.com]
> In detail, this task covers the following improvements for spark matrix right
> indexing. Frames are not covered here because they allow variable-length
> blocks. Also, note that it is important to differentiate between in-core and
> out-of-core matrices: for in-core matrices (i.e., matrices that fit in
> deserialized form into aggregated memory), the full scan is actually not
> problematic as the filter operation only scans keys without touching the
> actual values.
> (1) Scan-based indexing w/o aggregation: So far, we apply aggregations to
> merge partial blocks very conservatively. However, if the indexing range is
> block aligned (e.g., dimension start at block boundary or range within single
> block) this is unnecessary. This alone led to a 2x improvement for indexing
> row batches out of an in-core matrix.
> (2) Single-block lookup: If the indexing range covers a subrange of a single
> block, we directly perform a lookup. On in-core matrices this gives a minor
> improvement (but does not hurt) while on out-of-core matrices, the
> improvement is huge in case of existing partitioner as we only have to scan a
> single partition instead of the entire data.
> (3) Multi-block lookups: Unfortunately, Spark does not provide a lookup for a
> list of keys. So the next best option is a data-query join (in case of
> existing partitioner) with {{data.join(filter).map()}}, which works very well
> for in-core data sets, but for out-of-core datasets, unfortunately, does not
> exploit the potential for partition pruning and thus reads the entire data. I
> also experimented with a custom multi-block lookup that runs multiple lookups
> in a multi-threaded fashion - this gave the expected pruning but was very
> ugly due to an unbounded number of jobs.
> In conclusion, I'll create a patch for scenarios (1) and (2), while scenario
> (3) requires some more thoughts and is postponed after the 0.11 release. One
> idea would be to create a custom RDD that implements {{lookup(List<T> keys)}}
> by constructing a pruned set of input partitions via
> {{partitioner.getPartition(key)}}. cc [~freiss] [~niketanpansare] [~reinwald]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
