Retrieving the results_df on a crashed ray tune run

Ray Libraries (Data, Train, Tune, Serve) Ray Tune
1

Hi, newbie question here

I am running ray tune for some hyperparameter optimization. Everything I’ve done is on my local machine. After about 15 hours my optimization experiment crashed. My code did not save the results_df object (from analysis) because the exception occured inside tune.run.

The documentation online suggests that you can rebuild the results dataframe after the fact using the checkpoint files. However when I tried the examples, my code fails.
I’ve tried following the examples on this page, using the Analysis and ExperimentalAnalysis classes:

xa = ExperimentAnalysis(experiment_checkpoint_path=os.path.expanduser("~/ray_results/ax/my_checkpoint.json"))
xa.results_df
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/luke/.virtualenvs/myenv/lib/python3.8/site-packages/ray/tune/analysis/experiment_analysis.py", line 500, in results_df
    [
TypeError: 'NoneType' object is not iterable

a = Analysis(os.path.expanduser("~/ray_results/ax/"))
a.results_df
Traceback (most recent call last):
  File "<string>", line 1, in <module>
AttributeError: 'Analysis' object has no attribute 'results_df'

What am I doing wrong here?

Thanks