DEEP LEARNING EXPERIMENT TOOLKIT
The first real-time collaborative desktop app to support the full deep learning experiment life cycle.
Analysis tools for an experiment are a very important aspect of diagnosing defects or misbehavior of the actual deep learning model. Deepkit provides the first interactive Tensorflow debugger that allows to see into your model in real-time - simply integrated in your code using Keras Callbacks.
On top of that you have the option to display additional metrics such as accuracy, loss, text logs, debug images, or custom plots as well as debugging information about your model onto the experiment dashboard.
With Deepkit you can run any kind of script and have the ability to supervise, track, and version control every aspect of all of your machine learning experiments. While running your experiments, all of your used hyperparameters, environment, hardware utilization, execution times and results, as well as any kind of artifacts are all automatically saved and displayed for your convenience.
With integrated real-time collaboration tools using a Deepkit team server, you can work easily in a team and share not only your results, but also notes, source code and your computation servers. Deepkit supports any framework and any language, as long as it runs on Docker.
Connect any personal HPC device to the app or to your self-hosted centralized Deepkit team server and switch between them seamlessly. Deepkit allows you to group your different computation servers into clusters, which allows you to easily configure and change server access for different teams or team members.
Deepkit tracks and schedules your experiments according to your needs and quota settings. In future releases Deepkit will also allow you to configure directly in the app AWS or Google Cloud servers, making the process of going into the cloud super easy.