search

JerryI / wolfram-js-frontend

Dynamic Notebook Environment for Wolfram Language written in Javascript
https://jerryi.github.io/wljs-docs/
GNU General Public License v3.0
232 stars 8 forks source link

Support for Neural Networks output #186

Open s4m13337 opened 1 month ago

s4m13337 commented 1 month ago

It would be interesting to have support for neural network structure/graph output. Currently it just displays a tiny blank box.

Also when training the network with, no helpful information is displayed regarding the current status of training. If TrainingProgressReporting -> "Print" is set, then the information appears on the information bubble on the top right, which is not very useful. It would be easier to track if it gets printed on to the cell.

I hope this gets implemented in the near future as this would really accelerate my research work with neural networks.

JerryI commented 2 weeks ago

We tried our best, however, the developers of WL encoded the output form in somewhat complicated way, so that we are not able to overwrite the definitions of StandardForm for things such as LinearLayer and etc. Another option, we can still introduce a wrapper expressions for each problematic symbol. It wont' affect the evaluation process and UX, but will help to get informative output (this is how Graphics, Graph, ByteArray were implemented, where we failed to overwrite the definitions). Writing an output wrapper symbol is a bit cumbersome, therefore we ask for your help.

Could, you, please mention the most essential or required for your typical neural network workflow object symbols? There are too many of them to check. Ones I found are used the most (represent the state)