Thanks to the HN mods for letting me repost this. I'm very keen on hearing your thoughts on this tool and how I could make it better (especially if you do research in DNNs).

I'll follow up soon with a more interesting LSTM layer visualization.

Looks very useful -- I thought the video demo showing the example visualizations was very helpful in understanding what it does, and made it more approachable. Thank you for sharing!

This is hopefully a first step towards more understandable deep learning networks. Great job!

This looks great. I'm doing something similar, albeit not real time when learning about DNN right now and it helps hugely to understand what's going on.

Will definitely try it in the coming weeks. Thanks !

Whoa that's pretty cool! I've been meaning to check out TensorFlow and this may help me better understand how it works. Thanks!

This is completely awesome.

How do you do the side-by-side notebook and display thing Jupyter? Is it the new Jupyter dashboard/widget thing?

I use jquery to inject an entry div, then load React from there. I looked into ipywidgets, but it looks like that can only render one widget per cell. Jupyter Comm channels are used for communication with the kernel. Jupyter is an incredibly well-designed piece of software, which makes it pretty easy to extend the browser environment to do anything :)