Jupyter IFrame support

[radiantone]

Hi,
I have a keras model (tensorflow backend). Can I use this to visualize it in jupyter?

[lutzroeder]

See #96. Feedback is welcome.

[radiantone]

I dont think that #96 ticket provides all the details needed in this ticket.

How to export the keras model to browse?
What is the server port of the netron server?
I'm not running it locally.

Just need a few concrete details here. Thanks

[lutzroeder]

keras_model.save('test.h5')
!pip install netron
import netron
netron.start('test.h5', 8081)

[radiantone]

Thanks!

[sjain-stanford]

@lutzroeder has there been a change in API on latest release? I get this when I specify port here:

netron.start(model_path, port=8080)

TypeError: start() got an unexpected keyword argument 'port'

UPDATE: Never mind, just checked the source and address seems to be the argument to use instead of port.

[alexanderganderson]

It is possible to view netron in a jupyter notebook via an IFrame html element.

import IPython

import netron
port = 6006
model_path = # <Your favorite model>
netron.start(model_path, 6006, browse=False)

IPython.display.IFrame(f"http://localhost:{port}", width=1000, height=1000)

I'm running everything locally, so I'm not sure if you need to enable another port in your case.

[lutzroeder]

@alexanderganderson can you investigate and submit a pull request.

[lovettchris]

netron.start('model.onnx', 8081) works great the first time, and auto-opens my web browser, which is fine, but the second time I do this in Jupyter, it spends a very long time saying Stopping http://localhost:8081/. Can it "share" the server so if it's started once then it just serves up the next page instantly?

(I'm on Windows 11 doing this in VS Code version of Jupyter).

PS: I'd rather have the diagram show up inline inside the notebook. IPython.display.IFrame works, but is not pretty because browser is also launched and I still have the netron.start problem above.

[avilum]

https://gist.github.com/avilum/c4c595e972ba56384973d901addd3877