Is the model prediction probability calibrated?

[Vedant-R]

I understand that the sklearn random forest models prediction probabilities are not calibrated and we need to add steps in between to calibrate it.

Just wanted to understand if the prediction probabilities of tensorflow decision forests are calibrated?

If not, how can be calibrate it?

[janpfeifer]

No, they are not. But not by much usually -- depending on your use case it may not matter.

The extended discussion in TF Forum here, for others following up on this.

Leaving this issue as an enhancement to remind us to add support for automatic calibration.