AMLTKΒΆ

To be able to load your AMLTK run into DeepCave, there are a few points to consider when running AMLTK:

  1. Save the trial_history as a Parquet file:

    history_df = trial_history.df()
bucket.store({"history.parquet": history_df})

  2. Save the configuration space as a ConfigSpace JSON file:

    space = pipeline.search_space(parser="configspace")
bucket.store({"configspace.json": space.to_serialized_dict()})

  3. Define the start and end times for your trials:

    • DeepCAVE needs to know which columns to use as the trial start end end times. Ensure the start and end times are named in the format "deepcave:time:start" and "deepcave:time:end" in the history.parquet.

    • You can define these times during the AMLTK run setup. For example, assuming you have a data-loading and a scoring step and tracked their times via the profiler, to use the start time from data-loading and the end time of scoring as DeepCAVE start end end time, you can add the following lines before calling bucket.store() in step 1:

      history_df["deepcave:time:start"] = history_df["profile:data-loading:time:start"]
history_df["deepcave:time:end"] = history_df["profile:scoring:time:start"]

    • Alternatively, you can still manually add the DeepCave time columns to history.parquet after the AMLTK run has finished by loading it into a Pandas Dataframe, manipulating it, and writing it back to the history.parquet file.

previous

SMAC

next

Optuna