DeepCAVE

DeepCAVE is a visualization and analysis tool for AutoML, with a particular focus on hyperparameter optimization (HPO). Built on the Dash framework, it offers a fully interactive experience. The tool features a variety of plugins that enable efficient insight generation, aiding in understanding and debugging the application of HPO. Additionally, the powerful run interface and the modularized plugin structure allow extending the tool at any time effortlessly.

Installation

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

Afterwards, follow the instructions to install DeepCAVE:

conda create -n DeepCAVE python=3.9 conda activate DeepCAVE conda install -c anaconda swig pip install DeepCAVE

To load runs created with Optuna, the BOHB optimizer or RayTune, you need to install the respective packages by running:

pip install deepcave[optuna] pip install deepcave[bohb] pip install deepcave[raytune]

To try the examples for recording your results in DeepCAVE format, run this after installing:

pip install deepcave[examples]

If you want to contribute to DeepCAVE, use the following steps instead:

git clone https://github.com/automl/DeepCAVE.git cd DeepCAVE conda create -n DeepCAVE python=3.9 conda activate DeepCAVE conda install -c anaconda swig make install-dev

Please visit the documentation to get further help (e.g. if you cannot install redis server or if you are on MacOS).

Visualizing and Evaluating

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

deepcave --open

If you specify --open your webbrowser automatically opens at http://127.0.0.1:8050/. You can find more arguments and information (like using custom configurations) in the documentation.

Example runs

DeepCAVE comes with some pre-evaluated runs to get a feeling for what DeepCAVE can do.

If you cloned the repository from GitHub via git clone https://github.com/automl/DeepCAVE.git, you can try out some examples by exploring the logs directory inside the DeepCAVE dashboard. For example, if you navigate to logs/DeepCAVE/mnist_pytorch, you can view its runs if you hit the + button left to them.

Features

Interactive Interface

• Interactive Dashboard:

The dashboard runs in a webbrowser and allows you to self-analyze your optimization runs interactively.

• Run Selection Interface:

Easily select runs from your working directory directly within the interface.

• Integrated Help and Documentation:

Use help buttons and integrated documentation within the interface to better understand the plugins.

Comprehensive Analysis Tools

• Extensive Plugin Collection:

Explore a wide range of plugins for in-depth performance, hyperparameter, and budget analysis.

• Analysis of Running Processes:

Analyze and monitor optimization processes as they occur, with automatic detection of run changes.

• Group Analysis:

Choose groups of runs for combined analysis to gain deeper insights.

Flexible and Modular Architecture

• Modular Plugin Architecture:

Benefit from a modularized plugin structure with access to selected runs and groups, offering you maximum flexibility.

• Asynchronous Execution:

Utilize asynchronous execution of resource-intensive plugins and caching of results to improve performance.

Broad Optimizer Support

• Optimizer Support:

Work with many frameworks and optimizers using our converters, including converters for SMAC, BOHB, AMLTK, and Optuna.

• Native Format Saving:

Save AutoML runs from various frameworks in DeepCAVE's native format using the built-in recorder.

• Flexible Data Loading:

Alternatively, load AutoML runs from other frameworks by converting them into a Pandas DataFrame.

Developer and API Features

• API Mode:

Interact with the code directly through API mode, allowing you to bypass the graphical interface if preferred.

Citation

If you use DeepCAVE in one of your research projects, please cite our ReALML@ICML'22 workshop paper:

@misc{sass-realml2022,     title = {DeepCAVE: An Interactive Analysis Tool for Automated Machine Learning},     author = {Sass, René and Bergman, Eddie and Biedenkapp, André and Hutter, Frank and Lindauer, Marius},     doi = {10.48550/ARXIV.2206.03493},     url = {https://arxiv.org/abs/2206.03493},     publisher = {arXiv},     year = {2022},     copyright = {arXiv.org perpetual, non-exclusive license} }

Copyright (C) 2021-2024 The DeepCAVE Authors