I'm Martin Isaksson, the CEO & Co-founder of PerceptiLabs. We have developed a GUI and visual API for TensorFlow, on a mission to make ML modeling easier, faster, and more transparent.

While working with the beta version of TensorFlow back in 2016, my co-founder Robert Lundberg and I came up with the idea of PerceptiLabs. We felt that TensorFlow could benefit from improvements on the UX side because it was painful to see that such a powerful framework had so many UX challenges.

Today, TensorFlow has come a long way and is much easier to use, however, it has still a long road ahead of it to become easy enough to be used in an efficient way.

One aspect that Robert and I always felt was the most time-consuming part when developing ML models, was creating all the different kinds of plots and visualizations.

We created PerceptiLabs as a visual modeling tool, and here is a brief overview:

-Drag and drop interface for modelling, with warnings and recommendations

-I/O shape fitting

-Model debugging/interpretability: Auto-generates visualizations for all model variables both during modeling and training

-ML training infrastructure: Comes as a Docker image(s), distributes training on cluster(s)

-Model management

Since ML models are graphs, they are best suited to be represented as such. Visual graph representations are widely used (see papers at arXiv). This is why we believe that drag and drop functionality in conjunction with graph-oriented modeling, fits very well for ML – it just makes modeling more intuitive.

We recently launched a new version of PerceptiLabs: https://youtu.be/9QaiCjV2jK8

I hope that you will give it a try. The free version of PerceptiLabs can easily be installed and run:

$ pip install perceptilabs

$ perceptilabs

As we are trying to build the best and most intuitive ML modeling tool for you, we need your feedback in order to improve PerceptiLabs. We are nothing without our community :)