nengo.Node and learning rule don't work with the partitioner/assignments?

[adityagilra]

1

To get current nengo_mpi working with current nengo git repo '2.2.0.dev0', I had to do the following:

I need to comment in file "/root/nengo_git/nengo/cache.py", line 331:
#warnings.warn("Cannot shrink outside of a with cache block.")
Else a warning stops the run.

2

A minor issue:
I need to put in a line in simulator.py line 185: if len(data)>0:
(This is needed if I have a probe with a sample_every which is higher than dt, and is thus empty if a small run is given -- I give multiple run-s in a loop so over time it will not be empty.)

3

I am unable to get Node to work with the partitioner, it works (at least the script runs) without the partitioner.
Test script below:

import nengo_mpi
import nengo

model = nengo.Network()
with model:
    A = nengo.Node( size_in=2, size_out=2, output = lambda timeval,currval: (0.,0.) )
    #A = nengo.Ensemble(n_neurons=50, dimensions=2)
    B = nengo.Ensemble(n_neurons=50, dimensions=2)
    nengo.Connection(A, B)

#assignments = {A: 0, B: 1}
#sim = nengo_mpi.Simulator(model, assignments=assignments)
partitioner = nengo_mpi.Partitioner(n_components=8)
sim = nengo_mpi.Simulator(model, 1e-3, partitioner=partitioner)
sim.run(1.0)
sim.run(1.0)

4

With a learning rule on a connection (neurons to neurons), the partitioner gives this error:
(no error without the learning rule attached)
File "/root/nengo_mpi/nengo_mpi/partition/base.py", line 397, in merge_clusters
self._map[conn.learning_rule] = a
TypeError: unhashable type: 'dict'

and the assignments give this error:
nengo_mpi.PartitionError: Connection <Connection from <Neurons of <Ensemble (unlabeled) at 0x7ff20fad6350>> to <Neurons of <Ensemble (unlabeled) at 0x7ff228151450>>> crosses a component boundary, but it is not permitted to. Pre-object assigned to 0, post-object assigned to 1.

Script as below:

import nengo_mpi
import nengo

model = nengo.Network()
with model:
    #A = nengo.Node( size_in=2, size_out=2, output = lambda timeval,currval: (0.,0.) )
    A = nengo.Ensemble(n_neurons=50, dimensions=2)
    B = nengo.Ensemble(n_neurons=50, dimensions=2)
    AtoB = nengo.Connection(A.neurons, B.neurons)
    lruleDict = { 'PES' : nengo.PES(learning_rate=1e-3,pre_tau=1e-2) }
    AtoB.learning_rule_type = lruleDict

assignments = {A: 0, B: 1}
sim = nengo_mpi.Simulator(model, assignments=assignments)
#partitioner = nengo_mpi.Partitioner(n_components=8)
#sim = nengo_mpi.Simulator(model, 1e-3, partitioner=partitioner)
sim.run(1.0)
sim.run(1.0)

Thanks.