Github user vruusmann commented on the pull request:
https://github.com/apache/spark/pull/9207#issuecomment-218312022
@holdenk The JPMML-SparkML library depends on AGPLv3-licensed libraries,
which doesn't leave much choice.
I've just published the refactored feature mapping system. While
implementing support for the `StandardScalerModel` transformation type, I came
across a rather nasty issue related to the "tolerance" (actually, the lack
thereof) of continuous tree split conditions. For example, on several
occasions, the feature vector that was coming through `StandardScaler` was
dispatched to the wrong tree branch, because there was a difference in the
"16th decimal place of precision" between the computed feature value and the
threshold value (eg. `x = 0.36367792586219525` does not satisfy the split
condition `x <= 0.3636779258621952`).
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
