H2O v3.26.0.11 Release Notes
Release Date: 2019-05-12 // almost 5 years ago-
๐ Download at: http://h2o-release.s3.amazonaws.com/h2o/rel-yau/11/index.html
Bug
๐ [PUBDEV-6580] - The Python client now fails with descriptive message when attempting to run on an unsupported Java version. [PUBDEV-6895] - Fixed an issue that caused h2o to fail when running on Hadoop with
-internal_secure_connections
. [PUBDEV-6911] - H2OGenericEstimator can now be instantiated with no parameters. [PUBDEV-6945] - Multi-node H2O XGBoost now returns reproducible results. 0๏ธโฃ [PUBDEV-6995] - Fixed the backend default values for theinflection_point
andsmoothing
parameters in Target Encoder. [PUBDEV-7006] - Users can now specify thenoise
parameter when running Target Encoding in the R client or in Flow. โ [PUBDEV-7036] - MOJO reader now uses stderr instead of stdout to show warnings. ๐ [PUBDEV-7056] - Fixed an issue that allowed SPNEGO athentication to pass with any HTTP-Basic header. [PUBDEV-7062] - When connecting to H2O via the Python client, users can now specifyallowed_properties="cacert"
.New Feature
[PUBDEV-6213] - Added BroadcastJoinForTargetEncoding.
Task
[PUBDEV-6970] - Introduced AllCategorical and Threshold TE application strategies.
Improvement
โ [PUBDEV-7052] - Added a test to check XGBoost variable importance when trained on frames with shuffled input columns. ๐ฆ [PUBDEV-7053] - The package name for ai.h2o.org.eclipse.jetty.jaas.spi is now independent of the Jetty version. [PUBDEV-7060] - The
offset_column
is now propogated to MOJO models.๐ Docs
๐ [PUBDEV-7070] - Improved documentation for
stopping_metric
as it pertains to AutoML.