Request for -invert-hash type feature as in Vowpal Wabbit.

[musram]

Is your feature request related to a problem? Please describe. The getReadableModel() gives hash with the weight. The PR https://github.com/microsoft/SynapseML/pull/821 does refer to the requested feature but it was not committed.

Describe the solution you'd like We want feature: weight

Additional context Similar to -invert-hash of vowpal wabbit.

AB#1884838

[musram]

@eisber I was looking at your reply for https://github.com/microsoft/SynapseML/issues/1364.

I am finding it difficult to understand your statement "> Additionally it depends on the learning algo how the features are mapped to weights." If there are some additional transformations at the feature level happening within the VowpalWabbitRegressionModel, then it makes sense otherwise I am unable to understand the statement. Pls, correct me if I am wrong.

[eisber]

not all algos in VW have just a single weight for each feature.

@jackgerrits can help explain too.

[jackgerrits]

For example if using bagging, then internally VW maintains several models. Each occupying a different hash space. You need to know the number of bags in order to decipher what each hash may correspond to.