This demo implements an N-dimensional neural integrator.
This example utilizes a recurrent network. It shows how neurons can be used to implement stable dynamics. Such dynamics are important for memory, noise cleanup, statistical inference, and many other dynamic transformations.
It employs the EnsembleArray network, which provides a convenient method to act on multiple low-dimensional ensembles as though they were one high-dimensional ensemble.
[1]:
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import nengo
from nengo.networks import EnsembleArray
from nengo.processes import Piecewise
import nengo_loihi
nengo_loihi.set_defaults()
/home/travis/build/nengo/nengo-loihi/nengo_loihi/version.py:23: UserWarning: This version of `nengo_loihi` has not been tested with your `nengo` version (3.0.0.dev0). The latest fully supported version is 2.8.0
nengo.__version__, latest_nengo_version))
/home/travis/virtualenv/python3.5.2/lib/python3.5/site-packages/nengo_dl/version.py:32: UserWarning: This version of `nengo_dl` has not been tested with your `nengo` version (3.0.0.dev0). The latest fully supported version is 2.8.0.
((nengo.version.version,) + latest_nengo_version))
WARNING: Logging before flag parsing goes to stderr.
W0624 00:12:21.680078 140346077861696 deprecation_wrapper.py:119] From /home/travis/virtualenv/python3.5.2/lib/python3.5/site-packages/nengo_dl/__init__.py:55: The name tf.logging.set_verbosity is deprecated. Please use tf.compat.v1.logging.set_verbosity instead.
W0624 00:12:21.681391 140346077861696 deprecation_wrapper.py:119] From /home/travis/virtualenv/python3.5.2/lib/python3.5/site-packages/nengo_dl/__init__.py:55: The name tf.logging.WARN is deprecated. Please use tf.compat.v1.logging.WARN instead.
Our model consists of one recurrently connected ensemble array, and an input node for each dimension. The input nodes will provide piecewise step functions as input so that we can see the effects of recurrence.
[2]:
dimensions = 3
tau = 0.1
with nengo.Network(label='Integrator') as model:
ens = EnsembleArray(n_neurons=100, n_ensembles=dimensions)
stims = [
nengo.Node(
Piecewise({
0: 0,
0.2: np.random.uniform(low=0.5, high=3.),
1: 0,
2: np.random.uniform(low=-3, high=-0.5),
3: 0,
4: np.random.uniform(low=-3, high=3),
5: 0
})
)
for dim in range(dimensions)
]
for i, stim in enumerate(stims):
nengo.Connection(stim, ens.input[i], transform=[[tau]], synapse=tau)
# Connect the ensemble array to itself
nengo.Connection(ens.output, ens.input, synapse=tau)
# Collect data for plotting
stim_probes = [nengo.Probe(stim) for stim in stims]
ens_probe = nengo.Probe(ens.output, synapse=0.01)
We can use Nengo to see the desired model output.
[3]:
with nengo.Simulator(model) as sim:
sim.run(6)
t = sim.trange()
/home/travis/virtualenv/python3.5.2/lib/python3.5/site-packages/nengo/builder/optimizer.py:640: UserWarning: Skipping some optimization steps because SciPy is not installed. Installing SciPy may result in faster simulations.
warnings.warn("Skipping some optimization steps because SciPy is "
[4]:
def plot_decoded(t, data):
plt.figure()
for stim_probe in stim_probes:
plt.plot(t, data[stim_probe], color="b")
plt.plot(t, data[ens_probe], 'k')
plot_decoded(t, sim.data)
[5]:
with nengo_loihi.Simulator(model) as sim:
sim.run(6)
t = sim.trange()
/home/travis/build/nengo/nengo-loihi/nengo_loihi/discretize.py:468: UserWarning: Lost 1 extra bits in weight rounding
warnings.warn("Lost %d extra bits in weight rounding" % (-s2,))
/home/travis/build/nengo/nengo-loihi/nengo_loihi/discretize.py:468: UserWarning: Lost 2 extra bits in weight rounding
warnings.warn("Lost %d extra bits in weight rounding" % (-s2,))
[6]:
plot_decoded(t, sim.data)
