Justin Yu

The problem here is that the result dataframes get loaded from the `progress.csv` file by default. The CSV logger doesn't support adding new keys (and will just ignore them), and...

Hi @voklejas2, I reproduced the error with the script you provided. Here is how to fix the issue (I've also included an example of how to load the checkpoint): **Use...

Glad to hear! I made one small update to the checkpoint loading to use the recommended framework-specific [`TensorflowCheckpoint`](https://docs.ray.io/en/latest/ray-air/package-ref.html#ray.train.tensorflow.TensorflowCheckpoint) rather than loading it as a dict and getting the weights key...

@ju2ez @tmynn Just bumping this thread to see if any assistance is needed on my end! Changes look good and I think what's mostly left is to finish writing the...

Hi @asifmustafa87, I tested this on py38 and it seems to work with `ax-platform==0.3.2` and `typeguard==2.13.3`. Using py310 which is the default on Google Colab doesn't work for some reason....

@ramannanda9 What is the `FileSystemProvider` that you're mentioning?

@ramannanda9 Good to see that you found a solution. `pyarrow`'s default implementations do the job for most cases, and we felt that the flexibility of slotting in fsspec filesystems like...

@getorca You may want to call `result_grid.get_best_result(...).get_best_checkpoint(...)` to get the best checkpoint of the best trial. Otherwise, the default `checkpoint`/`metrics` attributes are just the latest as you mentioned.

> The best trial is properly returned from result._experiment_analysis.get_best_trial(metric='eval_accuracy', mode='max', scope='all') which is called before returning the object with the wrong checkpoint and accuracy from `_trial_to_result` Hmm, `result_grid.get_best_result` just calls...

`result.best_checkpoints` should return a list of all the top checkpoints for a trial across all iterations. Maybe I'm misunderstanding something?