freeze_params makes the network unpicklable about nengo-dl HOT 4 CLOSED

I'll take a look into the pickling behaviour, but if you want a way to do this for now that doesn't rely on pickling you can use the sim.save/load_params functions (which will save the parameters in a TensorFlow format, rather than through pickle). E.g.

net = build_network()
with nengo_dl.Simulator(net) as sim:
    sim.train(...)
    sim.save_params("my_params")

then to load them into a different backend

net = build_network()
with nengo_dl.Simulator(net) as sim:
    sim.load_params("my_params")
    sim.freeze_params()

with another_backend.Simulator(net) as sim:
    ...

from nengo-dl.

Oh neat. I was aware of those other methods but hadn't thought to combine them with the approach here. That should be good for now, thanks!

from nengo-dl.

Might be fixed in nengo/nengo#1584?

from nengo-dl.

Confirmed that this is resolved by the above fix in nengo core. That change hasn't yet made its way into a nengo core release, so anyone running into this issue should check out the development version of nengo core (3.1.0).

from nengo-dl.