« Changelog History


H2O v3.34.0.1 Release Notes

  • 🚀 Download at: http://h2o-release.s3.amazonaws.com/h2o/rel-zizler/1/index.html

    🐛 Bug

    • 🛠 [PUBDEV-8326] - Fixed matplotlib 3.4 compatibility issues with partial_plot.
    • 🗄 [PUBDEV-8316] - Deprecated is_supervised parameter for h2o.grid method in R.
    • 🛠 [PUBDEV-8314] - Fixed AutoML NPE by ensuring that models without metrics are not added to the leaderboard.
    • [PUBDEV-8295] - Redistributed the time budget for AutoML.
    • 🛠 [PUBDEV-8290] - Fixed and reorganized the H2O Explain leaderboard and fixed the confusion matrix.
    • [PUBDEV-8289] - Decreased the number of displayed features in the heatmap for AutoML inside H2O Explain.
    • 🛠 [PUBDEV-8276] - Fixed NPE raised from weight_column not being in the training model.
    • 📚 [PUBDEV-8274] - Fixed the weight=0 documentation change error.
    • ✅ [PUBDEV-8271] - Fixed failing rotterdam tests.
    • 🛠 [PUBDEV-8267] - Fixed GAM NPE from multiple runs with knots specified in a frame.
    • [PUBDEV-8266] - Fixed col_sample_rate not sampling for XGBoost when set to a value lower than 1.0.
    • 🛠 [PUBDEV-8257] - Fixed wrong column type on MOJO models for Cross-Validation Metrics Summary.
    • [PUBDEV-8245] - Prevented R connect from starting H2O locally.
    • [PUBDEV-8233] - Added StackedEnsembles to AutoML’s time budget to prevent unexpected training times.
    • [PUBDEV-8210] - Fixed the failing pyunit_scale_pca_rf.py test.
    • [PUBDEV-8175] - Improved AutoML behavior when multiple instances are created in parallel.
    • [PUBDEV-7855] - Solved corner cases involving mapping between encoded varimps and predictor columns for H2O Explain by making the varimp feature consolidation more robust.

    👌 Improvement

    • [PUBDEV-8273] - Ensured that AutoML uses the entire time budget for max_runtime.
    • [PUBDEV-8196] - Implemented custom progress widgets for Wave apps using H2O-3.
    • 🖨 [PUBDEV-8189] - Allowed users to convert floats to doubles with PrintMojo to prevent possible parsing issues.
    • ⚡️ [PUBDEV-8185] - Updated GBM cross validation with early_stopping to use ntrees that produce the best score.
    • 🖨 [PUBDEV-8184] - Enabled print_mojo to produce .png outputs.
    • ⚡️ [PUBDEV-8180] - Updated Python API for all algorithms and AutoML to retrieve the trained model or leader.
    • 🚚 [PUBDEV-8174] - Removed algorithm-specific logic from base classes.
    • 👍 [PUBDEV-8172] - Added support for scoreContributions for imported MOJOs in Java.
    • [PUBDEV-8170] - Exposed AutoML args as writeable properties until first called to train.
    • ⚡️ [PUBDEV-8168] - Updated XGBoost print_mojo to now output weights.
    • 🚚 [PUBDEV-8152] - Removed the Python client dependency on colorama.
    • [PUBDEV-8146] - Added the parameters and their default values to the _init_ function of the Py code generator.
    • [PUBDEV-8114] - Reduced the workspace of the validation frame in GBM by sharing it with the training frame in cross validation.
    • [PUBDEV-8085] - Slightly reduced precision of predictions stored in holdout frames to significantly save on memory.
    • 🚚 [PUBDEV-8015] - Removed warning in the Stacked Ensemble prediction function about missing fold_column frame.
    • [PUBDEV-7958] - Enabled returning data from Explain’s varimp_heatmap and model_correlation_matrix.
    • [PUBDEV-7937] - Exposed the top n and bottom n reason codes in Python/R and MOJO.
    • 🏗 [PUBDEV-5300] - Fixed nightly build version mismatch that prevented the H2OCluster timezone being set to America/Denver.

    🆕 New Feature

    • ✅ [PUBDEV-8319] - Implemented a java-self-check to allow users to run on latest Java.
    • ⚡️ [PUBDEV-8312] - Sped up GBM by optimizing the building of histograms.
    • ⚡️ [PUBDEV-8287] - Added a warning to the TreeSHAP reweighting feature if there are 0 weights and updated the API.
    • [PUBDEV-8235] - Added Maximum R Square Improvement (MAXR) algorithm to GLM.
    • ⚠ [PUBDEV-8229] - Added warning for when H2O doesn’t have enough memory to run XGBoost.
    • [PUBDEV-8221] - Added the ability to specify a custom file name when saving a MOJO.
    • 🖨 [PUBDEV-8203] - Added output version number of genmodel.jar when printing usage for PrintMojo.
    • [PUBDEV-8113] - Added MOJO to Rulefit.
    • [PUBDEV-8099] - Implemented ability to calculate Shapley values on a re-weighted tree.
    • [PUBDEV-8088] - Implemented H2O ANOVA GLM algorithm for GLM.
    • [PUBDEV-7354] - Improved and consolidated the handling of version mismatch between Python and Backend.
    • [PUBDEV-7139] - Implemented permutation feature importance for black-box models.
    • [PUBDEV-7138] - Implemented Extended Isolation Forest algorithm.
    • 👍 [PUBDEV-6364] - Added support for saving a model directly to S3.

    Task

    • 🔀 [PUBDEV-8292] - Fixed the time limits for the Merge/Sort benchmark.
    • 🚚 [PUBDEV-8197] - Switched removed pandas as_matrix method to .values and exposed the interim pandas.DataFrame object.
    • [PUBDEV-8116] - Fixed S3 credential for pyunit_s3_model_save.py test.
    • [PUBDEV-8084] - Connected XGBoost aggregation functionality with sorting functionality.

    Technical task

    • [PUBDEV-8202] - Replaced subsampling in Extended Isolation Forest.

    📄 Docs

    • ⚡️ [PUBDEV-8307] - Updated the AutoML FAQ.
    • [PUBDEV-8304] - Corrected the ignored_columns example.
    • 📚 [PUBDEV-8299] - Added RMarkdown, Jupyter Notebook, and HTML output example files to H2O Explain documentation.
    • 📚 [PUBDEV-8282] - Added Maximum R Improvements (MAXR) GLM documentation.
    • [PUBDEV-8261] - Added the loss function equations for each distribution and link type.
    • 📚 [PUBDEV-8248] - Updated the documentation about StackedEnsembles time constraints in AutoML.
    • [PUBDEV-8205] - Clarified that the Explain function only works for supervised models.
    • 📚 [PUBDEV-8179] - Added Examine Models section to AutoML documentation.
    • 📚 [PUBDEV-8166] - Added documentation for H2O ANOVA GLM algorithm.
    • 📚 [PUBDEV-8123] - Fixed the H2O Explain example in the documentation.
    • ⚡️ [PUBDEV-8053] - Updated and gathered Java links to a singular place in the User Guide.