Thanks for the tip, this is truly valuable!
My use case: I am plotting progress of Flux models learning in real time, e.g. plotting training & validation set loss, epoch time etc. So yes, it is development visualisation. The plot updates after every epoch (usually 0.5-10 seconds).
I also need those WebIO callbacks to work because I am doing stuff like
on(plt["relayout"]) do data
if data isa Dict && haskey(data, "dragmode")
plt.plot.layout.fields[:dragmode] = data["dragmode"]
The problems with Electron window are:
- They die when particular learning Julia REPL is closed. No way I know of to prevent that. I need the training reports to be more durable as I use them to compare with previous model’s etc. performance.
- It’s kind of impractical to switch between tens of those windows, while in browser they are nicely organised in tabs.
- They get in the way every time I start new learning process, while browser just opens new tab.