Philip May
Philip May
Hi @dmatrix - maybe you could provide some feedback about this FR?
Hello? Any news about this?
Hey dear project maintainers @harupy and @smurching : What about this suggestion / improvement. For me my colleagues and others this might be an important improvement of MLflow. What do...
Hi @dmatrix - maybe you could provide some feedback about this FR?
Hello? Any news about this?
Yes. learning_rate is not conditioned. I switched to this sampler: `sampler = optuna.samplers.CmaEsSampler(restart_strategy='ipop')` And get this warning: ``` [W 2020-09-23 09:39:55,691] The parameter 'learning_rate' in trial#53 is sampled independently by...
@hvy is there anything I can help you with to debug this?
All complete trials have LR set. There is one failed and one running trial stored in the DB which has `NaN` stored in LR.
@hvy yes. I think that should simulate what I have in my Optuna DB when this issue appears. But as you said. Everything that is not complete should be ignored....
@hvy I exported the DB out of the cloud environment for you: [optuna_copy.zip](https://github.com/optuna/optuna/files/5297210/optuna_copy.zip) ```python study_name='copy' storage='sqlite:///optuna_copy.db' study = optuna.create_study(study_name=study_name, storage=storage, load_if_exists=True, direction='maximize') df = study.trials_dataframe() print(df) fig = optuna.visualization.plot_param_importances(study) fig.write_html("../fig/optuna_plot_importances_{}.html".format(study_name))...