def test_SaveLoad():
""" Test save and load RecAEIFNest """
from rockpool.layers import RecIAFSpkInNest
from rockpool.layers import RecAEIFSpkInNest
from rockpool import timeseries as ts
from rockpool.networks import network as nw
import numpy as np
# - Generic parameters
weights_in = np.array([[-0.001, 0.002, 0.004], [0.03, -0.003, -0.0015]])
weights_rec = np.random.rand(3, 3) * 0.001
bias = 0.0
tau_mem = [0.02, 0.05, 0.1]
tau_syn_exc = [0.2, 0.01, 0.01]
tau_syn_inh = tau_syn_exc
dt = 0.001
vThresh = -0.055
vRest = -0.065
# - Layer generation
fl0 = RecAEIFSpkInNest(
weights_in=weights_in,
weights_rec=weights_rec,
dt=dt,
bias=bias,
tau_mem=tau_mem,
tau_syn_exc=tau_syn_exc,
tau_syn_inh=tau_syn_inh,
v_thresh=vThresh,
v_reset=vRest,
v_rest=vRest,
subthresh_adapt=0.0,
spike_adapt=0.001,
delta_t=0.0,
refractory=0.001,
record=True,
name="lyrNest",
)
net0 = nw.Network(fl0)
net0.save("tmp.model")
net1 = nw.Network.load("tmp.model")
# - Input signal
spikeTimes = np.arange(0, 1, dt)
spikeTrain0 = np.sort(np.random.choice(spikeTimes, size=20, replace=False))
channels = np.random.choice([0, 1], 20, replace=True).astype(int)
tsInEvent = ts.TSEvent(spikeTrain0, channels)
# - Compare states before and after
assert (np.abs(net0.input_layer.state - net1.input_layer.state) < 0.00001).all()
dFl0 = net0.evolve(tsInEvent, duration=1.0)
dFl1 = net1.evolve(tsInEvent, duration=1.0)
assert (np.abs(net0.input_layer.state - net1.input_layer.state) < 0.00001).all()
fl0.terminate()
net1.lyrNest.terminate()
def test_delays():
""" test delays """
from rockpool import timeseries as ts
from rockpool.layers import FFIAFNest, RecIAFSpkInNest
from rockpool.networks import network as nw
import numpy as np
weights = [[0.001, 0.0, 0.0, 0.0]]
bias = 0.375
tau_mem = 0.01
vReset = -0.07
vRest = -0.07
vTh = -0.055
fC = 0.25
dt = 0.001
refractory = 0.002
fl0 = FFIAFNest(
weights=weights,
dt=dt,
bias=bias,
tau_mem=tau_mem,
v_reset=vReset,
v_rest=vRest,
v_thresh=vTh,
capacity=fC,
refractory=refractory,
record=True,
name="FF",
)
weights_in = [
[0.015, 0.015, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0],
]
weights_rec = [
[0.0, 0.0, 0.015, 0.015],
[0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0014],
]
delay_in = [
[0.001, 0.011, 0.001, 0.001],
[0.001, 0.001, 0.001, 0.001],
[0.001, 0.001, 0.001, 0.001],
[0.001, 0.001, 0.001, 0.001],
]
delay_rec = [
[0.001, 0.001, 0.001, 0.011],
[0.001, 0.001, 0.001, 0.001],
[0.001, 0.001, 0.001, 0.001],
[0.001, 0.001, 0.001, 0.001],
]
vfBiasRec = 0.0
vtTauNRec = [0.2, 0.2, 0.2, 0.2]
tau_syn_exc_rec = [0.2, 0.2, 0.2, 0.2]
tau_syn_inh_rec = tau_syn_exc_rec
# - Layer generation
fl1 = RecIAFSpkInNest(
weights_in=weights_in,
weights_rec=weights_rec,
delay_in=delay_in,
delay_rec=delay_rec,
dt=0.001,
bias=vfBiasRec,
tau_mem=vtTauNRec,
capacity=0.1,
tau_syn_exc=tau_syn_exc_rec,
tau_syn_inh=tau_syn_inh_rec,
refractory=0.001,
record=True,
name="Rec",
)
net = nw.Network(fl0, fl1)
# - Input signal
vTime = np.arange(0, 1, dt)
vVal = np.zeros([len(vTime), 1])
vVal[500] = 0.01
tsInCont = ts.TSContinuous(vTime, vVal)
dAct = net.evolve(tsInCont, duration=1.0)
times = dAct["Rec"].times
eps = 0.000001
assert times[1] - times[0] - 0.01 < eps
assert times[3] - times[2] - 0.01 < eps
fl0.terminate()
fl1.terminate()
def test_timeconstants():
""" test delays """
from rockpool import timeseries as ts
from rockpool.layers import FFIAFNest, RecIAFSpkInNest
from rockpool.networks import network as nw
import numpy as np
weights = [[0.001]]
bias = 0.375
tau_mem = 0.01
vReset = -0.07
vRest = -0.07
vTh = -0.055
fC = 0.25
dt = 0.001
refractory = 0.002
fl0 = FFIAFNest(
weights=weights,
dt=dt,
bias=bias,
tau_mem=tau_mem,
v_reset=vReset,
v_rest=vRest,
v_thresh=vTh,
capacity=fC,
refractory=refractory,
record=True,
name="FF",
)
weights_in = [[0.001, -0.001]]
weights_rec = [[0, 0], [0, 0]]
vfBiasRec = 0.0
vtTauNRec = [0.2, 0.2]
tau_syn_exc_rec = [0.1, 0.1]
tau_syn_inh_rec = [0.01, 0.01]
# - Layer generation
fl1 = RecIAFSpkInNest(
weights_in=weights_in,
weights_rec=weights_rec,
dt=0.001,
bias=vfBiasRec,
tau_mem=vtTauNRec,
tau_syn_exc=tau_syn_exc_rec,
tau_syn_inh=tau_syn_inh_rec,
refractory=0.001,
v_reset=vReset,
v_rest=vRest,
v_thresh=vTh,
record=True,
name="Rec",
)
net = nw.Network(fl0, fl1)
# - Input signal
vTime = np.arange(0, 1, dt)
vVal = np.zeros([len(vTime), 1])
vVal[500] = 0.01
tsInCont = ts.TSContinuous(vTime, vVal)
dAct = net.evolve(tsInCont, duration=1.0)
exc_input = np.abs(fl1.recorded_states.samples[:, 0] - vRest)
inh_input = np.abs(fl1.recorded_states.samples[:, 1] - vRest)
# excitatory input peak should be later than inhibitory as the synaptic TC is longer
assert np.argmax(exc_input) > np.argmax(inh_input)
fl0.terminate()
fl1.terminate()
def test_DefaultParams():
""" Test RecIAFNest"""
from rockpool.layers import RecIAFSpkInNest, FFIAFNest
from rockpool import timeseries as ts
from rockpool.networks import network as nw
import numpy as np
# - Generic parameters
weights = np.ones([1, 2]) * 0.01
weights_in = np.array([[-0.1, 0.02, 0.4], [0.2, -0.3, -0.15]])
weights_rec = np.random.rand(3, 3) * 0.01
bias = 0.0
dt = 0.0001
tau_mem = 0.02
tau_syn_exc = 0.05
tau_syn_inh = 0.05
v_thresh = -0.055
v_reset = -0.065
v_rest = -0.065
capacity = tau_mem
refractory = 0.001
fl0 = FFIAFNest(
weights=weights,
dt=dt,
bias=bias,
tau_mem=tau_mem,
v_reset=v_reset,
v_rest=v_rest,
v_thresh=v_thresh,
capacity=capacity,
refractory=refractory,
record=True,
name="FF",
)
fl1 = RecIAFSpkInNest(
weights_in=weights_in,
weights_rec=weights_rec,
dt=dt,
bias=bias,
tau_mem=tau_mem,
tau_syn_exc=tau_syn_exc,
tau_syn_inh=tau_syn_inh,
v_thresh=v_thresh,
v_reset=v_reset,
v_rest=v_rest,
capacity=capacity,
refractory=refractory,
record=True,
name="Rec",
)
net0 = nw.Network(fl0, fl1)
fl2 = FFIAFNest(weights=weights, record=True, name="FF")
fl3 = RecIAFSpkInNest(
weights_in=weights_in, weights_rec=weights_rec, record=True, name="Rec"
)
net1 = nw.Network(fl2, fl3)
# - Input signal
vTime = np.arange(0, 1, dt)
vVal = np.zeros([len(vTime), 1])
vVal[2000:7000] = 0.01
tsInCont = ts.TSContinuous(vTime, vVal)
eps = 1e-6
assert (np.abs(fl0.state - fl2.state) < eps).all()
assert (np.abs(fl1.state - fl3.state) < eps).all()
# - Compare states before and after
dAct0 = net0.evolve(tsInCont, duration=1.0)
dAct1 = net1.evolve(tsInCont, duration=1.0)
assert (np.abs(fl0.state - fl2.state) < eps).all()
assert (np.abs(fl1.state - fl3.state) < eps).all()
fl0.terminate()
fl1.terminate()
def test_FFToRecLayerRepeat():
""" Test FFToRecNest"""
from rockpool import timeseries as ts
from rockpool.layers import FFIAFNest, RecIAFSpkInNest
from rockpool.networks import network as nw
import numpy as np
weights = [[0.0, 0.001, 0.0]]
bias = 0.375
tau_mem = 0.01
vReset = -0.07
vRest = -0.07
vTh = -0.055
fC = 0.25
dt = 0.001
refractory = 0.002
fl0 = FFIAFNest(
weights=weights,
dt=dt,
bias=bias,
tau_mem=tau_mem,
v_reset=vReset,
v_rest=vRest,
v_thresh=vTh,
capacity=fC,
refractory=refractory,
record=True,
name="FF",
)
weights_in = [[0.0, 0.0, 0.0], [0.0, 0.0, 0.6], [0.0, 0.0, 0.0]]
weights_rec = np.random.rand(3, 3) * 0.001
vfBiasRec = 0.0
vtTauNRec = [0.02, 0.05, 0.1]
tau_syn_exc_rec = [0.2, 0.01, 0.01]
tau_syn_inh_rec = tau_syn_exc_rec
# - Layer generation
fl1 = RecIAFSpkInNest(
weights_in=weights_in,
weights_rec=weights_rec,
dt=0.001,
bias=vfBiasRec,
tau_mem=vtTauNRec,
tau_syn_exc=tau_syn_exc_rec,
tau_syn_inh=tau_syn_inh_rec,
refractory=0.001,
record=True,
name="Rec",
)
net = nw.Network(fl0, fl1)
# - Input signal
vTime = np.arange(0, 1, dt)
vVal = np.zeros([len(vTime), 1])
vVal[500] = 0.01
tsInCont = ts.TSContinuous(vTime, vVal)
# - Compare states before and after
vStateBefore = np.copy(fl1.state)
for _ in range(10):
dAct = net.evolve(tsInCont, duration=1.0 / 10)
assert fl0.t == 1.0
assert fl1.t == 1.0
assert (vStateBefore != fl1.state).any()
net.reset_all()
assert fl0.t == 0
assert fl1.t == 0
assert (vStateBefore == fl1.state).all()
fl0.terminate()
def test_save_load():
""" Test RecIAFNest"""
from rockpool.layers import RecIAFSpkInNest, FFIAFNest, FFExpSynTorch
from rockpool import timeseries as ts
from rockpool.networks import network as nw
from rockpool.networks import Network as nws
import numpy as np
import pylab as plt
# - Generic parameters
weights = np.ones([1, 1]) * 0.01
weights_in = [[0.1, 0, 0]]
weights_rec = [[0, 0.1, 0], [0, 0, 0.1], [0.0, 0, 0]]
mfWOut = [[1], [1], [1]]
bias = 0.0
dt = 0.001
tau_mem = 0.02
tau_syn = 0.05
v_thresh = -0.055
v_reset = -0.065
v_rest = -0.065
capacity = 100.0
refractory = 0.001
np.random.seed(0)
fl0 = FFIAFNest(
weights=weights,
dt=dt,
bias=bias,
tau_mem=tau_mem,
v_reset=v_reset,
v_rest=v_rest,
v_thresh=v_thresh,
capacity=capacity,
refractory=refractory,
num_cores=1,
record=True,
name="FF",
)
fl1 = RecIAFSpkInNest(
weights_in=weights_in,
weights_rec=weights_rec,
dt=dt,
bias=bias,
tau_mem=tau_mem,
tau_syn_exc=tau_syn,
tau_syn_inh=tau_syn,
v_thresh=v_thresh,
v_reset=v_reset,
v_rest=v_rest,
capacity=capacity,
refractory=refractory,
num_cores=1,
record=True,
name="Rec",
)
net0 = nw.Network(fl0, fl1)
net0.save("test_nw.json")
net1 = nws.load("test_nw.json")
fl2 = net1.FF
fl3 = net1.Rec
np.random.seed(0)
# - Input signal
vTime = np.arange(0, 1, dt)
vVal = np.zeros([len(vTime), 1])
vVal[200:201] = 0.25
tsInCont = ts.TSContinuous(vTime, vVal)
epsilon = 0.0000001
# - Compare states before and after
np.random.seed(0)
dAct0 = net0.evolve(tsInCont, duration=1.0)
np.random.seed(0)
dAct1 = net1.evolve(tsInCont, duration=1.0)
assert (
np.abs(fl0.recorded_states.samples - fl2.recorded_states.samples) < epsilon
).all()
assert (
np.abs(fl1.recorded_states.samples - fl3.recorded_states.samples) < epsilon
).all()