Inconsitent node split values for different tree dump formats (dot, text, json)


during some tree interpretation, I have encountered some inconsistencies.
We can get the tree metadata using get_dump() method. The dump can be ‘text’, ‘json’ or ‘dot’.
The node’s split values are different between ‘text’ and ‘dot’. ‘text’ values seems to be ‘rounded’.

Could this be a bug ?


This is actually okay, since the text dump rounds up thresholds for binary variables.

thanks @hcho3, I would expect to have consistent values for text, json and dot dumps. When I saw it for the first time, I thought it is a strange bug. After more investigations, I realized the values could be rounded.
Do you consider that this need to be fixed?


No, I think this is fine.