ExperimentAnalysis object cannot be regenerated or unpickled

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

I’m using tune.run() to do some hypertuning on a model (with a Pytorch backend, using both a search algorithm (hyperopt) and a scheduler (ASHA)), and I’m finding that the ExperimentAnalysis object that tune.run() returns cannot be unpickled:

>>> exper = pickle.load(open("/data/ray_tune_analysis.pkl", "rb"))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "[...directories...]/python3.6/site-packages/ray/tune/trial.py", line 649, in __setstate__
    for key in self._nonjson_fields:
AttributeError: 'Trial' object has no attribute '_nonjson_fields'
>>>

…But the object created by pointing the constructor of ExperimentAnalysis at the experiment_state json never gets populated with a trials field, which makes the object of very limited usefulness:

>>> expa = ExperimentAnalysis(experiment_checkpoint_path="~/ray_results/SAC/experiment_state-2020-12-29_12-13-16.json")
No `self.trials`. Drawing logdirs from checkpoint file. This may result in some information that is out of sync, as checkpointing is periodic.
>>> expa.get_best_trial(metric="episode_reward_mean", mode="max")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "[...directories...]/python3.6/site-packages/ray/tune/analysis/experiment_analysis.py", line 535, in get_best_trial
for trial in self.trials:
TypeError: 'NoneType' object is not iterable

I can print the returned values of get_best_*() methods to stdout from the ExperimentAnalysis object returned by tune.run(), but without a way to save or regenerate those results, stdout seems like a bad option to settle on.

Is there a way to serialize the analysis object, or make it regenerate properly from the experiment directory?