Visualise textual Tensorflow summaries as a directed graph.
Find a file
2023-03-04 22:11:16 +00:00
lantern-build-engine@094e175740 add lantern 2023-02-16 22:24:42 +00:00
samples add sample summaries 2023-02-16 20:34:59 +00:00
src esbuild: add watching 2023-03-04 22:11:16 +00:00
.gitignore Initial commit 2023-02-16 19:05:43 +00:00
.gitmodules add lantern 2023-02-16 22:24:42 +00:00
build.sh build.sh: write initial script for CI 2023-02-16 22:30:48 +00:00
favicon.png add favicon 2023-02-18 02:53:38 +00:00
favicon.svg add favicon 2023-02-18 02:53:38 +00:00
LICENSE Initial commit 2023-02-16 20:17:05 +00:00
package-lock.json Initial commit 2023-02-16 19:05:43 +00:00
package.json esbuild: add watching 2023-03-04 22:11:16 +00:00
README.md tweak readme 2023-02-18 02:56:33 +00:00

tfsummaryvis

Visualise textual Tensorflow summaries as a directed graph.

Tensorflow summaries are invaluable to understand how a model is put together. They usually look like this:

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 bidirectional (Bidirectiona  (None, 100, 256)         336896    
 l)                                                              
                                                                 
 bidirectional_1 (Bidirectio  (None, 256)              394240    
 nal)                                                            
                                                                 
 dense (Dense)               (None, 2)                 514       
                                                                 
=================================================================
Total params: 731,650
Trainable params: 731,650
Non-trainable params: 0
_________________________________________________________________

Pretty easy to understand, right? But what about this monster? Or this one (this one doesn't actually work, do not implement at home)?

Larger models (and those aren't just a simple stack of layers) can be a nightmare to understand - even with a summary - hence this project was born :-)

It converts the above textual tensorflow summary into a directed graph with nomnoml.

Demo

Try the demo here: https://starbeamrainbowlabs.com/labs/tfsummaryvis/

Getting started locally

System Requirements

  • git
  • Node + npm

Instructions

If you want to run this project locally, first start by cloning the repository:

git clone https://git.starbeamrainbowlabs.com/sbrl/tfsummaryvis.git
cd tfsummaryvis;

Then, install dependencies:

npm install

Next, build the client-side app:

npm run build

Finally, serve with your favourite static web server:

npx serve dist/

Contributing

Contributions are very welcome - both issues and pull requests! Please mention in your pull request that you release your work under the MPL-2.0 (see below).

If you're feeling that way inclined, the sponsor button at the top of the page (if you're on GitHub) will take you to my Liberapay profile if you'd like to donate to say an extra thank you :-)

License

tfsummaryvis is released under the Mozilla Public License 2.0. The full license text is included in the LICENSE file in this repository. Tldr legal have a great summary of the license if you're interested.