Floating point imprecision

[edwardwliu]

When passing a data frame with 0 in training, the resulting processed_x occasionally replaces 0 with 1.387e-17. This is
likely due to floating point behavior either in the C++ logic or wrapper APIs. This has also led to flakiness in some exact tests.

[edwardwliu]

It's possible this is related to scaling per Theo

[petrovicboban]

@edwardwliu what's the status of this?

[edwardwliu]

Another issue that may be related to precision is that when predicting over the same dataset used in training, sometimes not all leaves are used.

There has been one case when honesty=TRUE and scale=TRUE where the number of unique leaves != number of unique predictions when predicting over the data used for averaging