zhengruifeng commented on a change in pull request #26832: [SPARK-30202][WIP][ML][PYSPARK] impl QuantileTransform URL: https://github.com/apache/spark/pull/26832#discussion_r367952349
########## File path: docs/ml-features.md ########## @@ -1495,6 +1495,57 @@ for more details on the API. </div> +## QuantileTransform + +QuantileTransform provide a non-parametric transformation to map the data to another +distribution, currently both uniform and gaussian are supported. +This model transforms the features to follow a uniform or a gaussian distribution. Review comment: Sorry for the late reply, recently I found some points on tree models seem more deserved to be improved in ml. > For ARIMA, this isn't how you'd make the data stationary; you're removing the distribution of the data. You'd use differencing? No, I have to take back my words. It is often desirable to transform a time series to make it stationary. Sometimes after applying Box-Cox with a appropriate value of _lambda_ the process may _look_ stationary. And this `QuantileTransform` can be to some degree viewed as a non-parameter variant of Box-Cox. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
