Error in deepcopy

hamed · July 27, 2020, 6:07pm

Hi,
I have a question and kindly need your help, How i can run a new and separate instance of current simulator at a specific state ?
It should be noted that, I have examined the following methods, but i haven’t succeed !!

1. copy.deepcopy has been tested and i have got errors about not having a Signal or Probe in Dict

simcopy=copy.deepcopy(sim)
simcopy.run_steps(5*1000)
t = simcopy.trange()
plot(simcopy)

github.com

hamed-Modiri/NengoTest/blob/71d2b2abe365c248e53807b354801246805f86c7/deepcopy.py#L170


    weights_p = nengo.Probe(conn, 'weights', synapse=0.01, sample_every=0.01)





sim = nengo.Simulator(model)

sim.run_steps(5*1000)

t = sim.trange()

plot(sim)



###############################################



simcopy=copy.deepcopy(sim)

simcopy.run_steps(5*1000)

t = simcopy.trange()

plot(simcopy)





###############################################



#state1=sim.__getstate__()



#sim0 = nengo.Simulator(model)

Or

state1=sim.getstate()
sim0 = nengo.Simulator(model)
sim0.setstate(copy.deepcopy(state1))
sim0.run_steps(5*1000)
plot(sim0)

2. also Pickle save and load has been used

with open(’.\sim3.dat’, ‘wb’) as f:
pickle.dump(sim, f)

with open(’.\sim3.dat’, ‘rb’) as f:
sim3=pickle.load(f)

sim3.run_steps(5*1000)

github.com

hamed-Modiri/NengoTest/blob/71d2b2abe365c248e53807b354801246805f86c7/deepcopy.py#L201


#with open('.\state1.dat', 'rb') as f:

#    state2==pickle.load(f)



#sim2 = nengo.Simulator(model)

#sim2.__setstate__(state2)

#sim2.run_steps(5*1000)

#plot(sim2)



###############################################



#with open('.\sim3.dat', 'wb') as f:

#    pickle.dump(sim, f)



#with open('.\sim3.dat', 'rb') as f:

#    sim3=pickle.load(f)



#sim3.run_steps(5*1000)





sim.run_steps(5*1000)

t = sim.trange()

Here is a simple script that reproduces error

github.com

hamed-Modiri/NengoTest/blob/master/deepcopy.py


import matplotlib
import matplotlib.pyplot as plt
matplotlib.use('TkAgg')
import nengo
import nengo_dl
from nengo.dists import Uniform
from nengo.utils.ensemble import response_curves
import numpy as np
import pickle
from nengo.solvers import LstsqL2
import MyLearning
import DecoderLearning
import RecurrentLearning
#import MyLearningDL
from clsgeneral import *
import copy


#--------------------------

This file has been truncated. show original

Any help will be greatly appreciated!

pblouw · July 28, 2020, 2:23am

Hi!
It is possible to pickle a closed Simulator object and unpickle it later on. However, once a simulator is closed, it can’t be reopened to resume the simulation (for some information on this topic, the following documentation might be useful: https://www.nengo.ai/nengo/backend-api.html#nengo.Simulator).

Depending on your use case, there are a couple of ways to simulate multiple copies of the same model. The most straightforward approach would involve setting a fixed seed to ensure deterministic behaviour across simulations. A more complicated approach would involve running a simulation to a current a given state, storing all of the state information, and then constructing a new network that uses this state information to initialize all of the e.g. neuron voltages, synaptic filters, etc. I wouldn’t recommend this unless there’s something specific you’re trying to do that isn’t handled by fixing the seed.

To demonstrate pickling and unpickling, the following script provides a short example:

import nengo
import numpy as np

with nengo.Network() as model:
    inp = nengo.Node(np.cos)
    ens = nengo.Ensemble(100, dimensions=1)
    nengo.Connection(inp, ens)
    probe = nengo.Probe(ens)


sim = nengo.Simulator(model)

with sim:
    sim.run(1)

with open('sim.pkl', 'wb') as pfile:
    pickle.dump(sim, pfile)


with open('sim.pkl', 'rb') as pfile:
    sim_copy = pickle.load(pfile)

Anyway, I hope this is helpful, but let us know if you have any followup questions!