DeepCAVE

DeepCAVE has two main contributions:

• Recording runs and
• Visualizing and evaluating trials of a run to get better insights into the AutoML process.

Installation

First, make sure you have swig and redis-server installed on your computer.

If you are on an Non-Intel Mac you have to add

export DISABLE_SPRING=true
export OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES

to your ~/.bash_profile to enable multi-processing.

Afterwards, follow the instructions:

git clone https://github.com/automl/DeepCAVE.git
cd DeepCAVE
conda env create -f environment.yml
conda activate DeepCAVE
make install

to your ~/.bash_profile to enable multi-processing.

If you want to contribute to DeepCAVE also install the dev packages:

make install-dev

Recording

In the following, a minimal example is given to show the simplicity yet powerful API to record runs.

import ConfigSpace as CS
from deep_cave import Recorder, Objective


configspace = CS.ConfigurationSpace(seed=0)
alpha = CS.hyperparameters.UniformFloatHyperparameter(
    name='alpha', lower=0, upper=1)
configspace.add_hyperparameter(alpha)

accuracy = Objective("accuracy", lower=0, upper=1, optimize="upper")
mse = Objective("mse", lower=0)

with Recorder(configspace, objectives=[accuracy, mse]) as r:
    for config in configspace.sample_configuration(100):
        for budget in [20, 40, 60]:
            r.start(config, budget)
            # Your code goes here
            r.end(costs=[0.5, 0.5])

Visualizing and Evaluating

The webserver as well as the queue/workers can be started by running

deepcave --start

Visit http://127.0.0.1:8050/ to get started. The following figures gives you a first impression of DeepCAVE. You can find more screenshots in the documentation.