3.2.0

• Fix bugs with distributed inference and evaluation
• Rename yaml related variables to json

3.0.0

• Update to support through the latest Tensorflow (2.6.0), which required converting YAML to JSON (https://github.com/tensorflow/tensorflow/releases/tag/v2.4.3).
• Remove hyperparameter optimization feature, as Hyperas is archived and causing some compatibility issues

2.1.0

• Add batch inference
• Provide capability to return training history
• Fix issue with saving Pipeline objects using ElephasEstimator stages

2.0.0

• Remove compatibility with Python 2.7
• Add support for Tensorflow 2.x +
• Remove custom predict logic
• Delete DL4J References

1.4.1

• Add support to predict probabilities in classification (#177)
• Fix gradient computation in synchronous mode (#176)
• Performance improvement for distributed predict (#176)

1.3.1

• Refactor transform method of ElephasTransform to perform prediction in a distributed manner

1.2.1

• Fix a minor bug with evaluate method of SparkModel
• Update some dependencies (remove six, change pyspark to < 3.2)

1.2.0

• Added support for custom objects (i.e; custom activation functions, custom layers, custom losses) in a model

1.1.0

• Add support for parallel prediction/evaluation in SparkModel
• Added backwards compatibility for ElephasEstimator with setFeaturesCol, setLabelCol, and setOutputCol, along with a deprecation warning
• Fixed a bug with socket server in hogwild mode (no longer behaving identical to asynchronous mode)
• Added typehints, as we're now only supporting Python 3

1.0.0

• Dropped support for Python 2.7
• Added support for Tensorflow 2.0.x - 2.1.x
• Added/fixed capability to train with socket client/server

Originally from: https://github.com/danielenricocahall/elephas/releases/tag/1.0.0