def plot_raster3d():
ax = Axes3D(fig)
y = np.array(spike_monitor.i) % num_neuron
z = np.array(spike_monitor.i) // num_neuron
x = spike_monitor.t / ms
ax.plot(x,
y,
z,
c="k",
marker=".",
markersize=2,
linewidth=0,
alpha=0.7)
ax.set_xlabel('Time')
ax.set_xlim(0, 500)
ax.set_ylabel('1st dimension')
ax.set_ylim(0, num_neuron)
ax.set_zlabel('2nd dimension')
ax.set_zlim(0, num_neuron)
ax.view_init(30, -70)
plt.savefig(num_trial + 'raster_2d.png', quality=95)
def test_correct_solution_2D(M):
"""
Function to test if solution is correct, returning detailed results.
M is a 2D matrix, i.e. cell-wise competition is assumed to have a winner.
"""
# Check if matrix is valid size.
M = np.array(M)
nrow, ncol = M.shape
if nrow != ncol:
raise ValueError('Matrix is not square size!')
if not np.sqrt(nrow).is_integer():
raise ValueError('Matrix length is not a square number!')
unique_vals = np.arange(nrow) + 1
# Test each row.
row_correct = np.array(
[np.array_equal(np.unique(M[i, :]), unique_vals) for i in range(nrow)])
# Test each column.
col_correct = np.array(
[np.array_equal(np.unique(M[:, j]), unique_vals) for j in range(ncol)])
# Test each sub-rectangle.
nsrow, nscol = int(np.sqrt(nrow)), int(np.sqrt(ncol))
sub_correct = [[
np.array_equal(
np.unique(M[(nsrow * i):(nsrow * (i + 1)),
(nscol * j):(nscol * (j + 1))]), unique_vals)
for j in range(nscol)
] for i in range(nsrow)]
sub_correct = np.array(sub_correct)
# Collest test results.
test_res = {'rows': row_correct, 'cols': col_correct, 'subs': sub_correct}
return test_res
def _extract_state_monitor(self, monitor):
self.f_set(record_variables=monitor.record_variables)
self.f_set(record=monitor.record)
self.f_set(when=monitor.when)
self.f_set(source=str(monitor.source))
self.f_set(name=monitor.name)
times = np.array(monitor.t[:])
if len(times) > 0:
self.f_set(t=times)
for varname in monitor.record_variables:
val = getattr(monitor, varname)
self.f_set(**{varname: val})
def _extract_state_monitor(self, monitor):
self.f_set(record_variables=monitor.record_variables)
self.f_set(record=monitor.record)
self.f_set(when=monitor.when)
self.f_set(source=str(monitor.source))
self.f_set(name=monitor.name)
times=np.array(monitor.t[:])
if len(times) > 0:
self.f_set(t=times)
for varname in monitor.record_variables:
val = getattr(monitor, varname)
self.f_set(**{varname: val})
def experimenter_bias(self):
""" Shows the statistics of the average recording across all experimenters """
# extracting rat id list
id_list = [rat_id for rat_id in self.data.data_vars]
# extracting experimenters name list - conversion for set and back to list for uniqueness
exp_name_list = [self.data[id].experimenter_name for id in id_list]
# extracting mean, average and std
descriptors = [(self.data[id].mean(), self.data[id].std(),
self.data[id].median()) for id in id_list]
# creating dataFrame variables
avrg = np.array(descriptors)[:, 0]
std = np.array(descriptors)[:, 1]
med = np.array(descriptors)[:, 2]
# dataframe instance
plt.figure()
df = pd.DataFrame({
'average': avrg,
'std': std,
'median': med
},
index=exp_name_list)
# barplot
df.plot.bar(rot=0)
plt.show()
f = h5py.File(STDP_WEIGHTS_FILE)
weights = f["weights"][:]
f.close()
e_synapses.w = weights * mV
else:
print("Initating new weights")
# adjust initiated weights to force output spike
e_synapses.w = 'rand()**4 * 2*mV'
# mon = StateMonitor(neurons, 'V', record=0)
spike_mon = SpikeMonitor(neurons)
run(sim_time, report='stdout')
weights = np.array(e_synapses.w / mV)
f = h5py.File(STDP_WEIGHTS_FILE, 'w')
f["weights"] = weights
f.close()
fig, ax = plt.subplots()
ax.hist(e_synapses.w / mV, bins=50)
ax.set(xlabel='$w_e$ (mV)')
fig, ax = plt.subplots()
ax.vlines(np.arange(n_repetitions * 10) * repeat_every / second,
0,
100,
color='gray',
alpha=0.5)
def plot_sudoku(M,
pM=None,
cM=None,
add_errors=None,
remove_lbls=True,
title=None,
fname=None):
"""
Plot Sudoku matrix, optionally adding errors.
M: a complete or partial solution.
pM: a partial solution, if provided, numbers are colored by differently.
cM: confidence matrix to scale size of numbers with.
"""
# Init.
M = np.array(M)
if pM is not None:
pM = np.array(pM)
if cM is not None:
cM = np.array(cM)
nrow, ncol = M.shape
nsrow, nscol = int(np.sqrt(nrow)), int(np.sqrt(ncol))
if add_errors is None:
# Add errors if matrix is complete.
add_errors = not np.any(np.isnan(M))
# Init figure.
base_cell_size = 1
ndigits_fac = 1 if nrow < 10 else 1.1
size = ndigits_fac * nrow * base_cell_size
fig = plt.figure(figsize=(size, size))
ax = plt.axes()
# Plot matrix.
sns.heatmap(M,
vmin=0,
vmax=0,
cmap='OrRd',
cbar=False,
square=True,
linecolor='k',
linewidth=1,
annot=False,
ax=ax)
# Add cell numbers.
for i, j in sudoku_util.all_ij_pairs(nrow):
lbl = int(M[i, j]) if not np.isnan(M[i, j]) else ''
# Color: is cell present in partial solution?
c = 'k' if pM is None else 'g' if not np.isnan(pM[i, j]) else 'b'
# Size: confidence level of cell.
s = 30 if cM is None else 10 + 20 * cM[i, j]
# Plot cell label.
ax.text(j + 0.5,
nrow - i - 0.5,
lbl,
va='center',
ha='center',
weight='bold',
fontsize=s,
color=c)
# Remove tick labels.
if remove_lbls:
ax.tick_params(labelbottom='off')
ax.tick_params(labelleft='off')
# Embolden border lines.
kws = {'linewidth': 6, 'color': 'k'}
for i in range(nsrow + 1):
ax.plot([0, ncol], [i * nsrow, i * nsrow], **kws)
for j in range(nscol + 1):
ax.plot([j * nscol, j * nscol], [0, ncol], **kws)
# Highlight errors.
if add_errors:
col, alpha = 'r', 1. / 3
test_res = sudoku_util.test_correct_solution_2D(M)
# Rows.
for i in np.where(np.logical_not(test_res['rows']))[0]:
irow = nrow - i - 1
rect = mpl.patches.Rectangle((0, irow),
ncol,
1,
alpha=alpha,
fc=col)
ax.add_patch(rect)
# Columns.
for j in np.where(np.logical_not(test_res['cols']))[0]:
rect = mpl.patches.Rectangle((j, 0), 1, nrow, alpha=alpha, fc=col)
ax.add_patch(rect)
# Sub-squares.
for i, j in np.argwhere(np.logical_not(test_res['subs'])):
isrow = nsrow - i - 1
rect = mpl.patches.Rectangle((j * nscol, isrow * nsrow),
nscol,
nsrow,
alpha=alpha,
fc=col)
ax.add_patch(rect)
# Add title.
if title is not None:
ax.set_title(title, fontsize='xx-large')
# Save figure.
if fname is not None:
fig.savefig(fname, dpi=300, bbox_inches='tight')
return ax
def __init__(
self,
weights: np.ndarray,
weights_in: np.ndarray,
dt: float = 0.1 * ms,
noise_std: float = 0 * mV,
refractory=0 * ms,
neuron_params=None,
syn_params=None,
integrator_name: str = "rk4",
name: str = "unnamed",
):
"""
Construct a spiking recurrent layer with IAF neurons, with a Brian2 back-end
:param weights: np.array NxN weight matrix
:param weights_in: np.array 1xN input weight matrix.
:param refractory: float Refractory period after each spike. Default: 0ms
:param neuron_params: dict Parameters to over overwriting neuron defaulst
:param syn_params: dict Parameters to over overwriting synapse defaulst
:param integrator_name: str Integrator to use for simulation. Default: 'exact'
:param name: str Name for the layer. Default: 'unnamed'
"""
warn("RecDynapseBrian: This layer is deprecated.")
# - Call super constructor
super().__init__(
weights=weights,
dt=np.asarray(dt),
noise_std=np.asarray(noise_std),
name=name,
)
# - Input weights must be provided
assert weights_in is not None, "weights_in must be provided."
# - Warn that nosie is not implemented
if noise_std != 0:
print("WARNING: Noise is currently not implemented in this layer.")
# - Set up spike source to receive spiking input
self._input_generator = b2.SpikeGeneratorGroup(
self.size, [0], [0 * second], dt=np.asarray(dt) * second)
# - Handle unit of dt: if no unit provided, assume it is in seconds
dt = np.asscalar(np.array(dt)) * second
### --- Neurons
# - Set up reservoir neurons
self._neuron_group = teiliNG(
N=self.size,
equation_builder=teiliDPIEqts(num_inputs=2),
name="reservoir_neurons",
refractory=refractory,
method=integrator_name,
dt=dt,
)
# - Overwrite default neuron parameters
if neuron_params is not None:
self._neuron_group.set_params(
dict(dTeiliNeuronParam, **neuron_params))
else:
self._neuron_group.set_params(dTeiliNeuronParam)
### --- Synapses
# - Add recurrent synapses (all-to-all)
self._rec_synapses = teiliSyn(
self._neuron_group,
self._neuron_group,
equation_builder=teiliDPISynEqts,
method=integrator_name,
dt=dt,
name="reservoir_recurrent_synapses",
)
self._rec_synapses.connect()
# - Add source -> reservoir synapses (one-to-one)
self._inp_synapses = teiliSyn(
self._input_generator,
self._neuron_group,
equation_builder=teiliDPISynEqts,
method=integrator_name,
dt=np.asarray(dt) * second,
name="receiver_synapses",
)
# Each spike generator neuron corresponds to one reservoir neuron
self._inp_synapses.connect("i==j")
# - Overwrite default synapse parameters
if syn_params is not None:
self._rec_synapses.set_params(neuron_params)
self._inp_synapses.set_params(neuron_params)
# - Add spike monitor to record layer outputs
self._spike_monitor = b2.SpikeMonitor(self._neuron_group,
record=True,
name="layer_spikes")
# - Call Network constructor
self._net = b2.Network(
self._neuron_group,
self._rec_synapses,
self._input_generator,
self._inp_synapses,
self._spike_monitor,
name="recurrent_spiking_layer",
)
# - Record neuron / synapse parameters
# automatically sets weights via setters
self.weights = weights
self.weights_in = weights_in
# - Store "reset" state
self._net.store("reset")
IMG_SIZE = 784
bin_size = 2 * ms
p = 1 * Hz * bin_size # Firing rate per neuron: 1Hz
total = 50 * second # Total length of our stimulus of one img
pattern_length = 250 * ms
repeat_every = pattern_length * 4
n_repetitions = int(total / repeat_every)
f = h5py.File(MNIST_TRAIN_HDF5_FILE, 'r')
train_img = f["img"][:]
train_label = f["label"][:]
f.close()
N = 100
global_indices = np.array([])
global_times = np.array([])
start_time = time.time()
for i in range(N):
spikes = np.random.rand(IMG_SIZE, int(total / bin_size)) < p
# create zero pattern first for we only fill nonzero position after
pattern = np.zeros([IMG_SIZE, int(pattern_length / bin_size)])
img_flatten = np.array(train_img[i]).flatten()
# This parameter is gotten using several experiments
rates = img_flatten / 255 * 125 * Hz
inp = PoissonGroup(IMG_SIZE, rates)
spikeMonitor = SpikeMonitor(inp)
