jmakov

That's also my experience. Catboost is one of the slowest even on multi GPU setup. Is this expected?

Not related to the issue itself, but wondering why you went with arrow dataset instead of parquet dataset.

Same problem here. Been online the whole time, no maintenance, dropped from 100% to 95%.

@Hoxbro thanks a lot for the quick feedback. OK, so it is doable :) . I've added the output of `.show_versions` to the original post. From a quick look, doesn't...

Tried numba 0.58.1, same exception

Using VS code and Jupyterlab produces this 

Same with `df[df.columns[0]].hvplot(datashade=True)` - memory increases on every execution (hvplot=0.7.3). Cannot use `%reset out` (causes an exception) because following https://docs.ray.io/en/latest/using-ray-with-jupyter.html?highlight=jupyter. conda's env.yaml: ```shell name: test channels: - pyviz - conda-forge...

Bump. I have to restart the kernel all the time in the notebook. Any suggestions very welcome (`gc.collect()` doesn't change anything).

@maximlt thanks for the quick response! I can reproduce this min example below in `jupyter-lab`: ```python import pandas pandas.options.plotting.backend = "holoviews" # `datashade=True` doesn't work without this line # I'm...

Would just like to bump this issue a bit since it's almost a blocker for me - imagine a long computation, then trying to plot with diff params, each plot...