Github user desperadoxx commented on the issue:
https://github.com/apache/spark/pull/18337
I also met this problem. The rmse doesn't get converged on some common
dataset, including movielens.
At least, it should not be clipped by maxValue and minValue when computing
the gradient, which will change loss function to non-convex, but the original
is convex.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
