Alongside regression, we'd also like to have classification analysis type where the output value is categorical rather than numerical. Steps: - [x] classification analysis type in C++ https://github.com/elastic/ml-cpp/pull/701 - [x] classification analysis type in Java https://github.com/elastic/elasticsearch/pull/46537 - [x] evaluation API https://github.com/elastic/elasticsearch/pull/46653 https://github.com/elastic/elasticsearch/pull/47126 - [x] HLRC https://github.com/elastic/elasticsearch/pull/46537 https://github.com/elastic/elasticsearch/pull/47126 - [x] HLRC docs and doc tests https://github.com/elastic/elasticsearch/pull/47569 Enhancements: - [x] support for boolean and integer types for dependent variable - [x] set the default for `num_top_classes` param to `2` - [x] make format of MulticlassConfusionMatrix metric result more understandable - [x] default "prediction_field_name" to (dependent_variable + "_prediction")
Alongside regression, we'd also like to have classification analysis type where the output value is categorical rather than numerical.
Steps:
Introduce classification analysis runner. ml-cpp#701
Implement new analysis type: classification #46537
Allow evaluation to consist of multiple steps. #46653
Implement evaluation API for multiclass classification problem #47126
Implement new analysis type: classification #46537
Implement evaluation API for multiclass classification problem #47126
Add MlClientDocumentationIT tests for classification. #47569
Enhancements:
num_top_classesparam to2