Tensorflow-decision-forests

Features

- Check learner parameters during the model construction.
- Fix discretized numerical features for regression task.
- Allow for float32 values to be fed as categorical features.
- Add new / improved tutorials for ranking and visualization.
- Compatibility with Tensorflow 2.12.0. Unfortunately, this means dropping
support for Python 3.7.

Fix

- Fix crashes when using ranking with very large groups.
- Add option to set the port used by YDF in TF-DF distributed training.
- Improve logging robustness.

Features

- Add support for distributed training and distributed hyper-parameter tuning
in the OSS build. See
https://www.tensorflow.org/decision_forests/distributed_training
- Setting "subsample" is enough enable random subsampling (to need to also set
"sampling_method=RANDOM").
- Add "min_vocab_frequency" argument in "FeatureUsage" to control the minimum
frequency of categorical items.
- Add "override_global_imputation_value" argument in "FeatureUsage" to
override the value used for global imputation of missing value by the
global-imputation algorithm.
- The Tuner argument "use_predefined_hps" automatically configures the set of
hyper-parameters to explore during automatic hyper-parameter tuning.
- Replaces the MEAN_MIN_DEPTH variable importance with INV_MEAN_MIN_DEPTH.
- Add option to forbid model inference with the slow inference engine.

Fix

- Automatic documentation generation for RandomForestModel and other classes.

Features

- Support for Tensorflow Serving APIs.
- Add support for zipped Yggdrasil Decision Forests model for
`yggdrasil_model_to_keras_model`.
- Added model prediction tutorial.
- Prevent premature stopping of GBT training through new parameter
`early_stopping_initial_iteration`.

Fix

- Using loaded datasets with TF-DF no longer fails (Github 131).
- Automatically infer the semantic of int8 values as numerical (was
categorical before).
- Build script fixed
- Model saving no longer fails when using invalid feature names.
- Added keyword to pandas dataset drop (Github 135).

Fix

- Issue in the application of auditwheel in TF 1.0.0.

Features

- Add customization of the number of IO threads when using
`fit_on_dataset_path`.

Fix

- Improved documentation
- Improved testing and stability

Features

- Multithreading of the oblique splitter for gradient boosted tree models.
- Support for pure serving model i.e. model containing only serving data.
- Add "edit_model" cli tool.

Fix

- Remove bias toward low outcome in uplift modeling.