Jupyter Notebook

CAVE is compatible with jupyter notebooks, that means all plots and tables will render in a jupyter notebook. There is an example in examples/cave_notebook.ipynb, that you can start with jupyter-notebook example/cave_notebook.ipynb. To use CAVE in such an interactive mode, you have to create a CAVE-instance:

from cave.cavefacade import CAVE
cave = CAVE(folders=["examples/smac3/example_output/run_1"],
            output_dir="test_jupyter_smac",
            ta_exec_dir=["examples/smac3"],
            show_jupyter=True,
           )

You can then use any of CAVE’s methods to analyze the loaded results by simply using it:

cave.parallel_coordinates()

This also holds for budget-based optimizers.

cave = CAVE(folders=["examples/bohb"],
            output_dir="test_jupyter_bohb",
            show_jupyter=True,
           )
cave.performance_table()