* rank_metric: add AUC-PR
Implementation of the AUC-PR calculation for weighted data, proposed by Keilwagen, Grosse and Grau (https://doi.org/10.1371/journal.pone.0092209)
* rank_metric: fix lint warnings
* Implement tests for AUC-PR and fix implementation
* add aucpr to documentation for other languages
* add back train method but mark as deprecated
* add back train method but mark as deprecated
* fix scalastyle error
* fix scalastyle error
* tiny fix for empty partition in predict
* further fix
* add back train method but mark as deprecated
* add back train method but mark as deprecated
* fix scalastyle error
* fix scalastyle error
* fix the pattern in dev script and version mismatch
* add back train method but mark as deprecated
* add back train method but mark as deprecated
* fix scalastyle error
* fix scalastyle error
* add dev script to update version and update versions
* add back train method but mark as deprecated
* add back train method but mark as deprecated
* fix scalastyle error
* fix scalastyle error
* update resource files
* Update SparkParallelismTracker.scala
* remove xgboost-tracker.properties
* [jvm-packages] Train Booster from an existing model
* Align Scala API with Java API
* Existing model should not load rabit checkpoint
* Address minor comments
* Implement saving temporary boosters and loading previous booster
* Add more unit tests for loadPrevBooster
* Add params to XGBoostEstimator
* (1) Move repartition out of the temp model saving loop (2) Address CR comments
* Catch a corner case of training next model with fewer rounds
* Address comments
* Refactor newly added methods into TmpBoosterManager
* Add two files which is missing in previous commit
* Rename TmpBooster to checkpoint
* [jvm-packages] Fixed test/train persistence
Prior to this patch both data sets were persisted in the same directory,
i.e. the test data replaced the training one which led to
* training on less data (since usually test < train) and
* test loss being exactly equal to the training loss.
Closes#2945.
* Cleanup file cache after the training
* Addressed review comments
* [jvm-packages] Exposed train-time evaluation metrics
They are accessible via 'XGBoostModel.summary'. The summary is not
serialized with the model and is only available after the training.
* Addressed review comments
* Extracted model-related tests into 'XGBoostModelSuite'
* Added tests for copying the 'XGBoostModel'
* [jvm-packages] Fixed a subtle bug in train/test split
Iterator.partition (naturally) assumes that the predicate is deterministic
but this is not the case for
r.nextDouble() <= trainTestRatio
therefore sometimes the DMatrix(...) call got a NoSuchElementException
and crashed the JVM due to lack of exception handling in
XGBoost4jCallbackDataIterNext.
* Make sure train/test objectives are different
In the refactor to add base margins, #2532, all of the labels were lost
when creating the dmatrix. This became obvious as metrics like ndcg
always returned 1.0 regardless of the results.
Change-Id: I88be047e1c108afba4784bd3d892bfc9edeabe55
Training a model with the experimental rank:ndcg objective incorrectly
returns a Classification model. Adjust the classification check to
not recognize rank:* objectives as classification.
While writing tests for isClassificationTask also turned up that
obj_type -> regression was incorrectly identified as a classification
task so the function was slightly adjusted to pass the new tests.
* Add SparkParallelismTracker to prevent job from hanging
* Code review comments
* Code Review Comments
* Fix unit tests
* Changes and unit test to catch the corner case.
* Update documentations
* Small improvements
* cancalAllJobs is problematic with scalatest. Remove it
* Code Review Comments
* Check number of executor cores beforehand, and throw exeception if any core is lost.
* Address CR Comments
* Add missing class
* Fix flaky unit test
* Address CR comments
* Remove redundant param for TaskFailedListener
* Allowed subsampling test from the training data frame/RDD
The implementation requires storing 1 - trainTestRatio points in memory
to make the sampling work.
An alternative approach would be to construct the full DMatrix and then
slice it deterministically into train/test. The peak memory consumption
of such scenario, however, is twice the dataset size.
* Removed duplication from 'XGBoost.train'
Scala callers can (and should) use names to supply a subset of
parameters. Method overloading is not required.
* Reuse XGBoost seed parameter to stabilize train/test splitting
* Added early stopping support to non-distributed XGBoost
Closes#1544
* Added early-stopping to distributed XGBoost
* Moved construction of 'watches' into a separate method
This commit also fixes the handling of 'baseMargin' which previously
was not added to the validation matrix.
* Addressed review comments
* Converted ml.dmlc.xgboost4j.LabeledPoint to Scala
This allows to easily integrate LabeledPoint with Spark DataFrame APIs,
which support encoding/decoding case classes out of the box. Alternative
solution would be to keep LabeledPoint in Java and make it a Bean by
generating boilerplate getters/setters. I have decided against that, even
thought the conversion in this PR implies a public API change.
I also had to remove the factory methods fromSparseVector and
fromDenseVector because a) they would need to be duplicated to support
overloaded calls with extra data (e.g. weight); and b) Scala would expose
them via mangled $.MODULE$ which looks ugly in Java.
Additionally, this commit makes it possible to switch to LabeledPoint in
all public APIs and effectively to pass initial margin/group as part of
the point. This seems to be the only reliable way of implementing distributed
learning with these data. Note that group size format used by single-node
XGBoost is not compatible with that scenario, since the partition split
could divide a group into two chunks.
* Switched to ml.dmlc.xgboost4j.LabeledPoint in RDD-based public APIs
Note that DataFrame-based and Flink APIs are not affected by this change.
* Removed baseMargin argument in favour of the LabeledPoint field
* Do a single pass over the partition in buildDistributedBoosters
Note that there is no formal guarantee that
val repartitioned = rdd.repartition(42)
repartitioned.zipPartitions(repartitioned.map(_ + 1)) { it1, it2, => ... }
would do a single shuffle, but in practice it seems to be always the case.
* Exposed baseMargin in DataFrame-based API
* Addressed review comments
* Pass baseMargin to XGBoost.trainWithDataFrame via params
* Reverted MLLabeledPoint in Spark APIs
As discussed, baseMargin would only be supported for DataFrame-based APIs.
* Cleaned up baseMargin tests
- Removed RDD-based test, since the option is no longer exposed via
public APIs
- Changed DataFrame-based one to check that adding a margin actually
affects the prediction
* Pleased Scalastyle
* Addressed more review comments
* Pleased scalastyle again
* Fixed XGBoost.fromBaseMarginsToArray
which always returned an array of NaNs even if base margin was not
specified. Surprisingly this only failed a few tests.
* Deduplicated DataFrame creation in XGBoostDFSuite
* Extracted dermatology.data into MultiClassification
* Moved cache cleaning to SharedSparkContext
Cache files are prefixed with appName therefore this seems to be just the
place to delete them.
* Removed redundant JMatrix calls in xgboost4j-spark
* Slightly more readable buildDenseRDD in XGBoostGeneralSuite
* Generalized train/test DataFrame construction in XGBoostDFSuite
* Changed SharedSparkContext to setup a new context per-test
Hence the new name: PerTestSparkSession :)
* Fused Utils into PerTestSparkSession
* Whitespace fix in XGBoostDFSuite
* Ensure SparkSession is always eagerly created in PerTestSparkSession
* Renamed PerTestSparkSession->PerTest
because it was doing slightly more than creating/stopping the session.
* [jvm-packages] Deduplicated train/test data access in tests
All datasets are now available via a unified API, e.g. Agaricus.test.
The only exception is the dermatology data which requires parsing a
CSV file.
* Inlined Utils.buildTrainingRDD
The default number of partitions for local mode is equal to the number
of available CPUs.
* Replaced dataset names with problem types
Prior to this commit XGBoostModel.predict produced an RDD with
an array of predictions for each partition, effectively changing
the shape wrt the input RDD. A more natural contract for prediction
API is that given an RDD it returns a new RDD with the same number
of elements. This allows the users to easily match inputs with
predictions.
This commit removes one layer of nesting in XGBoostModel.predict output.
Even though the change is clearly non-backward compatible, I still
think it is well justified. See discussion in 06bd5dca for motivation.
* Disabled excessive Spark logging in tests
* Fixed a singature of XGBoostModel.predict
Prior to this commit XGBoostModel.predict produced an RDD with
an array of predictions for each partition, effectively changing
the shape wrt the input RDD. A more natural contract for prediction
API is that given an RDD it returns a new RDD with the same number
of elements. This allows the users to easily match inputs with
predictions.
This commit removes one layer of nesting in XGBoostModel.predict output.
Even though the change is clearly non-backward compatible, I still
think it is well justified.
* Removed boxing in XGBoost.fromDenseToSparseLabeledPoints
* Inlined XGBoost.repartitionData
An if is more explicit than an opaque method name.
* Moved XGBoost.convertBoosterToXGBoostModel to XGBoostModel
* Check the input dimension in DMatrix.setBaseMargin
Prior to this commit providing an array of incorrect dimensions would
have resulted in memory corruption. Maybe backport this to C++?
* Reduced nesting in XGBoost.buildDistributedBoosters
* Ensured consistent naming of the params map
* Cleaned up DataBatch to make it easier to comprehend
* Made scalastyle happy
* Added baseMargin to XGBoost.train and trainWithRDD
* Deprecated XGBoost.train
It is ambiguous and work only for RDDs.
* Addressed review comments
* Revert "Fixed a singature of XGBoostModel.predict"
This reverts commit 06bd5dcae7780265dd57e93ed7d4135f4e78f9b4.
* Addressed more review comments
* Fixed NullPointerException in buildDistributedBoosters
When using xgboost4j-spark I had executors getting killed much more
often than i would expect by yarn for overrunning their memory limits,
based on the memoryOverhead provided. It looks like a significant
amount of this is because dmatrix's were being created but not released,
because they were only released when the GC decided it was time to
cleanup the references.
Rather than waiting for the GC, relesae the DMatrix's when we know
they are no longer necessary.
* add back train method but mark as deprecated
* fix scalastyle error
* fix the persistence of XGBoostEstimator
* test persistence of a complete pipeline
* fix compilation issue
* do not allow persist custom_eval and custom_obj
* fix the failed tesl
* add back train method but mark as deprecated
* fix scalastyle error
* change class to object in examples
* fix compilation error
* small fix for cleanExternalCache
* add back train method but mark as deprecated
* fix scalastyle error
* change class to object in examples
* fix compilation error
* fix several issues in tests
* [jvm-packages] call setGroup for ranking task
* passing groupData through xgBoostConfMap
* fix original comment position
* make groupData param
* remove groupData variable, use xgBoostConfMap directly
* set default groupData value
* add use groupData tests
* reduce rank-demo size
* use TaskContext.getPartitionId() instead of mapPartitionsWithIndex
* add DF use groupData test
* remove unused varable
* add back train method but mark as deprecated
* fix scalastyle error
* first commit in scala binding for fast histo
* java test
* add missed scala tests
* spark training
* add back train method but mark as deprecated
* fix scalastyle error
* local change
* first commit in scala binding for fast histo
* local change
* fix df frame test
* add back train method but mark as deprecated
* fix scalastyle error
* change class to object in examples
* fix compilation error
* bump spark version to 2.1
* preserve num_class issues
* fix failed test cases
* rivising
* add multi class test
* [jvm-packages] Scala implementation of the Rabit tracker.
A Scala implementation of RabitTracker that is interface-interchangable with the
Java implementation, ported from `tracker.py` in the
[dmlc-core project](https://github.com/dmlc/dmlc-core).
* [jvm-packages] Updated Akka dependency in pom.xml.
* Refactored the RabitTracker directory structure.
* Fixed premature stopping of connection handler.
Added a new finite state "AwaitingPortNumber" to explicitly wait for the
worker to send the port, and close the connection. Stopping the actor
prematurely sends a TCP RST to the worker, causing the worker to crash
on AssertionError.
* Added interface IRabitTracker so that user can switch implementations.
* Default timeout duration changes.
* Dependency for Akka tests.
* Removed the main function of RabitTracker.
* A skeleton for testing Akka-based Rabit tracker.
* waitFor() in RabitTracker no longer throws exceptions.
* Completed unit test for the 'start' command of Rabit tracker.
* Preliminary support for Rabit Allreduce via JNI (no prepare function support yet.)
* Fixed the default timeout duration.
* Use Java container to avoid serialization issues due to intermediate wrappers.
* Added tests for Allreduce/model training using Scala Rabit tracker.
* Added spill-over unit test for the Scala Rabit tracker.
* Fixed a typo.
* Overhaul of RabitTracker interface per code review.
- Removed methods start() waitFor() (no arguments) from IRabitTracker.
- The timeout in start(timeout) is now worker connection timeout, as tcp
socket binding timeout is less intuitive.
- Dropped time unit from start(...) and waitFor(...) methods; the default
time unit is millisecond.
- Moved random port number generation into the RabitTrackerHandler.
- Moved all Rabit-related classes to package ml.dmlc.xgboost4j.scala.rabit.
* More code refactoring and comments.
* Unified timeout constants. Readable tracker status code.
* Add comments to indicate that allReduce is for tests only. Removed all other variants.
* Removed unused imports.
* Simplified signatures of training methods.
- Moved TrackerConf into parameter map.
- Changed GeneralParams so that TrackerConf becomes a standalone parameter.
- Updated test cases accordingly.
* Changed monitoring strategies.
* Reverted monitoring changes.
* Update test case for Rabit AllReduce.
* Mix in UncaughtExceptionHandler into IRabitTracker to prevent tracker from hanging due to exceptions thrown by workers.
* More comprehensive test cases for exception handling and worker connection timeout.
* Handle executor loss due to unknown cause: the newly spawned executor will attempt to connect to the tracker. Interrupt tracker in such case.
* Per code-review, removed training timeout from TrackerConf. Timeout logic must be implemented explicitly and externally in the driver code.
* Reverted scalastyle-config changes.
* Visibility scope change. Interface tweaks.
* Use match pattern to handle tracker_conf parameter.
* Minor clarification in JNI code.
* Clearer intent in match pattern to suppress warnings.
* Removed Future from constructor. Block in start() and waitFor() instead.
* Revert inadvertent comment changes.
* Removed debugging information.
* Updated test cases that are a bit finicky.
* Added comments on the reasoning behind the unit tests for testing Rabit tracker robustness.
* add back train method but mark as deprecated
* fix scalastyle error
* change class to object in examples
* fix compilation error
* update methods in test cases to be consistent
* add blank lines
* fix
* add back train method but mark as deprecated
* fix scalastyle error
* change class to object in examples
* fix compilation error
* fix mis configuration
