def main(config):
TAU_F = config['TAU_F']
W_FF = config['W_FF']
V_TH = config['V_TH']
G = config['G']
V_0 = config['V_0']
DS = config['DS']
FONT_SIZE = config['FONT_SIZE']
FIG_SIZE = config['FIG_SIZE']
def f(v, d):
return (1 / TAU_F) * (-v + W_FF*phi(G*(v - V_TH)) + V_0 + d)
vs = np.linspace(-6, 10, 200)
fs = [f(vs, d) for d in DS]
fig, ax = plt.subplots(1, 1, figsize=FIG_SIZE)
ax.set_xlabel('v', fontsize=16)
ax.set_ylabel('f(v)', fontsize=16)
ax.plot(vs, np.transpose(fs), lw=2)
ax.legend(['d = {}'.format(d) for d in DS], loc='best')
ax.axvline(0, ls='--', lw=1, c='k')
ax.axhline(0, ls='--', lw=1, c='k')
axis_tools.set_fontsize(ax, FONT_SIZE)
def test3_input_populations_coupled_to_bistable_population(self):
# define network drive
drive_1 = 10*[0] + 20*[10] + 200*[0]
drive_2 = 15*[0] + 20*[10] + 195*[0]
drive_12 = len(drive_1) * [0]
drives = np.transpose([drive_1, drive_2, drive_12])
# define network parameters
taus = [10, 10, 10]
v_rests = [0, 0, 0]
v_ths = [4, 4, 4]
gs = [2.5, 2.5, 2.5]
noise_level = 2
nodes = [{'tau': t, 'v_rest': v_rest, 'threshold': th, 'steepness': g}
for t, v_rest, th, g in zip(taus, v_rests, v_ths, gs)]
weights = np.array([
[0, 0, 0],
[0, 0, 0],
[2, 2, 5.5]
])
# make network
ntwk = RateBasedModel(nodes, weights)
ntwk.store_voltages = True
ntwk.noise_level = noise_level
# initialize and run network
np.random.seed(1)
ntwk.vs = np.array(v_rests)
for drive in drives:
ntwk.step(drive)
fig, axs = plt.subplots(3, 1, figsize=(14, 8), sharex=True)
axs[0].plot(ntwk.vs_history, lw=2)
axs[1].plot(ntwk.rs_history, lw=2)
axs[2].plot(drives, lw=2)
axs[1].set_ylim(-.1, 1.1)
axs[2].set_ylim(-1, 11)
axs[0].set_ylabel('voltage')
axs[1].set_ylabel('firing rate')
axs[2].set_ylabel('drive')
axs[2].set_xlabel('t')
axs[0].set_title('Input populations coupled to bistable population')
for ax in axs:
axis_tools.set_fontsize(ax, 20)
def test2_upstate_timecourse_with_self_connections_depends_on_noise(self):
# define network drive
drives = 10*[np.array([0, 0, 0])] + 40*[4*np.array([1, 1, 1])] + \
140*[np.array([0, 0, 0])]
# define network parameters
taus = [10, 10, 10]
v_rests = [0, 0, 0]
v_ths = [4, 4, 4]
gs = [2.5, 2.5, 2.5]
w_selfs = [5.3, 5.3, 5.3]
noise_level = np.array([1, 2, 3])
nodes = [{'tau': t, 'v_rest': v_rest, 'threshold': th, 'steepness': g}
for t, v_rest, th, g in zip(taus, v_rests, v_ths, gs)]
weights = np.diag(w_selfs)
# make network
ntwk = RateBasedModel(nodes, weights)
ntwk.store_voltages = True
ntwk.noise_level = noise_level
# initialize and run network
np.random.seed(seed=4)
ntwk.vs = np.array(v_rests)
for drive in drives:
ntwk.step(drive)
# make plots
fig, axs = plt.subplots(3, 1, figsize=(14, 8), sharex=True)
axs[0].plot(ntwk.vs_history, lw=2)
axs[1].plot(ntwk.rs_history, lw=2)
axs[2].plot(drives, lw=2, c='k')
axs[1].set_ylim(-.1, 1.1)
axs[2].set_ylim(-1, 5)
axs[0].set_ylabel('voltage')
axs[1].set_ylabel('firing rate')
axs[2].set_ylabel('drive')
axs[2].set_xlabel('t')
axs[0].set_title('Self-Connected Populations With Noise')
for ax in axs:
axis_tools.set_fontsize(ax, 20)
def main():
# make network
nodes = [{'tau': TAU, 'v_rest': v_rest, 'threshold': V_TH, 'steepness': G} for v_rest in V_RESTS]
weights = np.array([
[0., 0, 0, 0, CWO],
[0, 0, 0, CWO, 0],
[FSW, FSW, SRW, 0, 0],
[CWI, 0, CGW, 0, 0],
[0, CWI, CGW, 0, 0],
])
ntwk = network.RateBasedModel(nodes, weights)
ntwk.store_voltages = True
ntwk.noise_level = NOISE_LEVEL
# set up network drive
drive_1 = 10*[0] + 20*[10] + 100*[0] + 20*[10] + 80*[0]
drive_2 = 25*[0] + 20*[10] + 185*[0]
drive_12s = len(drive_1) * [0]
drive_12 = len(drive_1) * [0]
drive_21 = len(drive_1) * [0]
drives = np.transpose([drive_1, drive_2, drive_12s, drive_12, drive_21])
t = np.arange(len(drives))
# run network
np.random.seed(SEED)
ntwk.vs = np.array(V_RESTS)
for drive in drives:
ntwk.step(drive)
# make figure
fig, axs = plt.subplots(3, 1, facecolor=FACE_COLOR, figsize=FIG_SIZE, sharex=True)
axs[0].plot(t, ntwk.vs_history, lw=2)
axs[1].plot(t, ntwk.rs_history, lw=2)
axs[2].plot(t, drives, lw=2)
axs[0].set_ylabel('voltage')
axs[1].set_ylabel('firing rate')
axs[2].set_ylabel('drive')
axs[2].set_xlabel('t')
axs[0].set_title('Two-component memory network')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
def test1_self_connections_yield_bistable_systems_given_correct_parameters(self):
# define network drive
drives = 10*[np.array([0, 0])] + 20*[4*np.array([1, 1])] + \
60*[np.array([0, 0])] + 20*[-4*np.array([1, 1])] + 60*[np.array([0, 0])]
# define network parameters
taus = [10, 10]
v_rests = [0, 0]
v_ths = [4, 4]
gs = [2.5, 2.5]
w_selfs = [7.5, 5.5]
nodes = [{'tau': t, 'v_rest': v_rest, 'threshold': th, 'steepness': g}
for t, v_rest, th, g in zip(taus, v_rests, v_ths, gs)]
weights = np.diag(w_selfs)
# make network
ntwk = RateBasedModel(nodes, weights)
ntwk.store_voltages = True
# initialize and run network
ntwk.vs = np.array(v_rests)
for drive in drives:
ntwk.step(drive)
# make plots
fig, axs = plt.subplots(3, 1, figsize=(14, 8), sharex=True)
axs[0].plot(ntwk.vs_history, lw=2)
axs[1].plot(ntwk.rs_history, lw=2)
axs[2].plot(drives, lw=2, c='k')
axs[1].set_ylim(-.1, 1.1)
axs[2].set_ylim(-5, 5)
axs[0].set_ylabel('voltage')
axs[1].set_ylabel('firing rate')
axs[2].set_ylabel('drive')
axs[2].set_xlabel('t')
axs[0].set_title('Self-Connected Populations')
for ax in axs:
axis_tools.set_fontsize(ax, 20)
def test0_unconnected_populations_with_different_timescales_respond_to_delta_pulse_with_exponential_decay(self):
# define network drive
drives = 10*[np.array([0, 0, 0, 0])] + 10*[2*np.array([3, 4, 6, 10])] + 60*[np.array([0, 0, 0, 0])]
# define network parameters
taus = [3, 5, 10, 20]
v_rests = [-20, -10, 0, 10]
v_ths = [-10, -1, 8, 17]
gs = [.5, .5, .5, .5]
# dake new network
nodes = [{'tau': t, 'v_rest': v_rest, 'threshold': th, 'steepness': g}
for t, v_rest, th, g in zip(taus, v_rests, v_ths, gs)]
weights = np.zeros((len(taus), len(taus)))
ntwk = RateBasedModel(nodes, weights)
ntwk.store_voltages = True
# initialize and run network
ntwk.vs = np.array(v_rests)
for drive in drives:
ntwk.step(drive)
# make plots
fig, axs = plt.subplots(3, 1, figsize=(14, 8), sharex=True)
axs[0].plot(ntwk.vs_history, lw=2)
axs[1].plot(ntwk.rs_history, lw=2)
axs[2].plot(drives, lw=2)
axs[1].set_ylim(-.1, 1.1)
axs[2].set_ylim(0, 45)
axs[0].set_ylabel('voltage')
axs[1].set_ylabel('firing rate')
axs[2].set_ylabel('drive')
axs[2].set_xlabel('t')
axs[0].set_title('Uncoupled Populations')
for ax in axs:
axis_tools.set_fontsize(ax, 20)
def novel_pattern_learning(CONFIG):
"""
Show that a network has an increased probability of embedding a novel pattern
into its connectivity network if we allow nonassociative priming to act.
"""
SEED = CONFIG["SEED"]
LOAD_FILE_NAME = CONFIG["LOAD_FILE_NAME"]
W_WEAK = CONFIG["W_WEAK"]
GAIN = CONFIG["GAIN"]
REFRACTORY_STRENGTH = CONFIG["REFRACTORY_STRENGTH"]
LINGERING_INPUT_VALUE = CONFIG["LINGERING_INPUT_VALUE"]
LINGERING_INPUT_TIMESCALE = CONFIG["LINGERING_INPUT_TIMESCALE"]
W_MAX = CONFIG["W_MAX"]
ALPHA = CONFIG["ALPHA"]
STRONG_DRIVE_AMPLITUDE = CONFIG["STRONG_DRIVE_AMPLITUDE"]
RUN_LENGTH = CONFIG["RUN_LENGTH"]
N_REPEATS = CONFIG["N_REPEATS"]
FIG_SIZE_0 = CONFIG["FIG_SIZE_0"]
FONT_SIZE = CONFIG["FONT_SIZE"]
np.random.seed(SEED)
# create new network with STDP learning rule
ntwk_old = np.load(LOAD_FILE_NAME)[0]
w = ntwk_old.w.copy()
# add weak connection to element 2 of node_1 path tree from element 1 of node_0 path tree
w[ntwk_old.node_1_path_tree[0][2], ntwk_old.node_0_path_tree[0][1]] = W_WEAK
w_to_track = (ntwk_old.node_1_path_tree[0][2], ntwk_old.node_0_path_tree[0][1])
path_novel = ntwk_old.node_0_path_tree[0][:2] + ntwk_old.node_1_path_tree[0][2:]
# make new base network
ntwk_base = network.RecurrentSoftMaxLingeringSTDPModelBasic(
w, GAIN, REFRACTORY_STRENGTH, LINGERING_INPUT_VALUE, LINGERING_INPUT_TIMESCALE, W_MAX, ALPHA
)
ntwk_base.node_0 = ntwk_old.node_0
ntwk_base.node_0 = ntwk_old.node_1
ntwk_base.node_0_path_tree = ntwk_old.node_0_path_tree
ntwk_base.node_1_path_tree = ntwk_old.node_1_path_tree
# show long trials of novel sequence drive, spontaneous activity, and test sequence
fig, axs = plt.subplots(N_REPEATS, 1, figsize=FIG_SIZE_0, sharex=True, tight_layout=True)
axs_twin = [ax.twinx() for ax in axs]
drives = np.zeros((RUN_LENGTH, ntwk_base.w.shape[0]), dtype=float)
for ctr, node in enumerate(path_novel):
drives[ctr, node] = STRONG_DRIVE_AMPLITUDE
for ax, ax_twin in zip(axs, axs_twin):
ntwk = deepcopy(ntwk_base)
ntwk.store_voltages = True
ws = []
for drive in drives:
ntwk.step(drive)
ws.append(ntwk.w[w_to_track])
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(ax, spikes, drives)
ax_twin.plot(ws, color="b", lw=2, alpha=0.7)
ax.set_ylabel("active \n ensemble")
ax_twin.set_ylabel("W({}, {})".format(*w_to_track), color="b")
axs[-1].set_xlabel("time step")
for ax_twin in axs_twin:
ax_twin.set_ylim(0, 2)
axs[0].set_xlim(-5, RUN_LENGTH)
axs[0].set_title("With nonassociative priming")
for ax in list(axs) + axs_twin:
axis_tools.set_fontsize(ax, FONT_SIZE)
fig, axs = plt.subplots(N_REPEATS, 1, figsize=FIG_SIZE_0, sharex=True, tight_layout=True)
axs_twin = [ax.twinx() for ax in axs]
drives = np.zeros((RUN_LENGTH, ntwk_base.w.shape[0]), dtype=float)
for ctr, node in enumerate(path_novel):
drives[ctr, node] = STRONG_DRIVE_AMPLITUDE
for ax, ax_twin in zip(axs, axs_twin):
ntwk = deepcopy(ntwk_base)
ntwk.lingering_input_value = 0
ntwk.store_voltages = True
ws = []
for drive in drives:
ntwk.step(drive)
ws.append(ntwk.w[w_to_track])
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(ax, spikes, drives)
ax_twin.plot(ws, color="b", lw=2, alpha=0.7)
ax.set_ylabel("active \n ensemble")
ax_twin.set_ylabel("W({}, {})".format(*w_to_track), color="b")
axs[-1].set_xlabel("time step")
for ax_twin in axs_twin:
ax_twin.set_ylim(0, 2)
axs[0].set_xlim(-5, RUN_LENGTH)
axs[0].set_title("Without nonassociative priming")
for ax in list(axs) + axs_twin:
axis_tools.set_fontsize(ax, FONT_SIZE)
def main(config):
SEED = config['SEED']
TAU_E = config['TAU_E']
TAU_I = config['TAU_I']
V_REST_E = config['V_REST_E']
V_REST_I = config['V_REST_I']
V_TH = config['V_TH']
STEEPNESS = config['STEEPNESS']
W_EE = config['W_EE']
W_EI = config['W_EI']
W_IE = config['W_IE']
NOISE_LEVEL = config['NOISE_LEVEL']
DRIVE_AMPS = config['DRIVE_AMPS']
DRIVE_STARTS = config['DRIVE_STARTS']
DRIVE_ENDS = config['DRIVE_ENDS']
DURATION = config['DURATION']
FIG_SIZE = config['FIG_SIZE']
FONT_SIZE = config['FONT_SIZE']
COLOR_CYCLE = config['COLOR_CYCLE']
# set up nodes and weights
nodes = [
{'tau': TAU_E, 'v_rest': V_REST_E, 'threshold': V_TH, 'steepness': STEEPNESS,},
{'tau': TAU_I, 'v_rest': V_REST_I, 'threshold': V_TH, 'steepness': STEEPNESS,},
]
weights = np.array([
[W_EE, W_EI],
[W_IE, 0.0],
])
# build network
ntwk = network.RateBasedModel(nodes, weights)
ntwk.store_voltages = True
ntwk.noise_level = NOISE_LEVEL
# set up drive
drives = np.zeros((DURATION, len(nodes)), dtype=float)
for drive_amp, drive_start, drive_end in zip(DRIVE_AMPS, DRIVE_STARTS, DRIVE_ENDS):
drives[drive_start:drive_end, 0] = drive_amp
# run network
np.random.seed(SEED)
ntwk.vs = np.array([V_REST_E, V_REST_I])
for drive in drives:
ntwk.step(drive)
# make figure
fig, axs = plt.subplots(3, 1, figsize=FIG_SIZE, sharex=True, tight_layout=True)
axs_twin = [ax.twinx() for ax in axs[:2]]
axs[0].plot(np.array(ntwk.vs_history)[:, 0], c=COLOR_CYCLE[0], ls='--', lw=2)
axs_twin[0].plot(np.array(ntwk.rs_history)[:, 0], c=COLOR_CYCLE[0], ls='-', lw=2)
axs[1].plot(np.array(ntwk.vs_history)[:, 1], c=COLOR_CYCLE[1], ls='--', lw=2)
axs_twin[1].plot(np.array(ntwk.rs_history)[:, 1], c=COLOR_CYCLE[1], ls='-', lw=2)
axs[2].set_color_cycle(COLOR_CYCLE)
axs[2].plot(drives, lw=2)
axs_twin[0].set_ylim(0, 1)
axs_twin[1].set_ylim(0, 1)
axs[2].set_ylim(0, drives.max() * 1.1)
axs[0].set_title('Excitatory')
axs[0].set_ylabel('Voltage')
axs_twin[0].set_ylabel('Firing rate')
axs[1].set_title('Inhibitory')
axs[1].set_ylabel('Voltage')
axs_twin[1].set_ylabel('Firing rate')
axs[2].set_title('Drive')
axs[2].set_ylabel('Drive')
axs[2].set_xlabel('t')
for ax in list(axs) + list(axs_twin):
axis_tools.set_fontsize(ax, FONT_SIZE)
def main(config):
# unpack params from config
SEED = config['SEED']
TAU_I = config['TAU_I']
TAU_R = config['TAU_R']
TAU_B = config['TAU_B']
TAU_A = config['TAU_A']
V_REST_I = config['V_REST_I']
V_REST_R = config['V_REST_R']
V_REST_B = config['V_REST_B']
V_REST_A = config['V_REST_A']
STEEPNESS = config['STEEPNESS']
THRESHOLD = config['THRESHOLD']
W_IB = config['W_IB']
W_BI = config['W_BI']
W_BR = config['W_BR']
W_AB = config['W_AB']
W_AA = config['W_AA']
NOISE_LEVEL = config['NOISE_LEVEL']
DURATION = config['DURATION']
DRIVE_11_START = config['DRIVE_11_START']
DRIVE_11_END = config['DRIVE_11_END']
DRIVE_11_AMP = config['DRIVE_11_AMP']
DRIVE_12_START = config['DRIVE_12_START']
DRIVE_12_END = config['DRIVE_12_END']
DRIVE_12_AMP = config['DRIVE_12_AMP']
DRIVE_13_START = config['DRIVE_13_START']
DRIVE_13_END = config['DRIVE_13_END']
DRIVE_13_AMP = config['DRIVE_13_AMP']
DRIVE_21_START = config['DRIVE_21_START']
DRIVE_21_END = config['DRIVE_21_END']
DRIVE_21_AMP = config['DRIVE_21_AMP']
COLOR_CYCLE = config['COLOR_CYCLE']
FIG_SIZE = config['FIG_SIZE']
FONT_SIZE = config['FONT_SIZE']
# set up nodes
# order: (I, R1, R2, R3, R4, B12, B13, B14, B23, B24, B34, A12, A13, A14, A23, A24, A34)
nodes = [
{'tau': TAU_I, 'v_rest': V_REST_I, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # I
{'tau': TAU_R, 'v_rest': V_REST_R, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # R1
{'tau': TAU_R, 'v_rest': V_REST_R, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # R2
{'tau': TAU_R, 'v_rest': V_REST_R, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # R3
{'tau': TAU_R, 'v_rest': V_REST_R, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # R4
{'tau': TAU_B, 'v_rest': V_REST_B, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # B12
{'tau': TAU_B, 'v_rest': V_REST_B, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # B13
{'tau': TAU_B, 'v_rest': V_REST_B, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # B14
{'tau': TAU_B, 'v_rest': V_REST_B, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # B23
{'tau': TAU_B, 'v_rest': V_REST_B, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # B24
{'tau': TAU_B, 'v_rest': V_REST_B, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # B34
{'tau': TAU_A, 'v_rest': V_REST_A, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # A12
{'tau': TAU_A, 'v_rest': V_REST_A, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # A13
{'tau': TAU_A, 'v_rest': V_REST_A, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # A14
{'tau': TAU_A, 'v_rest': V_REST_A, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # A23
{'tau': TAU_A, 'v_rest': V_REST_A, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # A24
{'tau': TAU_A, 'v_rest': V_REST_A, 'threshold': THRESHOLD, 'steepness': STEEPNESS}, # A34
]
# set up weight matrix
weights = np.array([
# I R1 R2 R3 R4 B12 B13 B14 B23 B24 B34 A12 A13 A14 A23 A24 A34
[ 0.0, 0.0, 0.0, 0.0, 0.0, W_IB, W_IB, W_IB, W_IB, W_IB, W_IB, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,], # I
[ 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, 0.0, 0.0, 0.0,], # R1
[ 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, 0.0, 0.0, 0.0,], # R2
[ 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, 0.0, 0.0, 0.0,], # R3
[ 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, 0.0, 0.0, 0.0,], # R4
[W_BI, W_BR, W_BR, 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,], # B12
[W_BI, W_BR, 0.0, W_BR, 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,], # B13
[W_BI, W_BR, 0.0, 0.0, W_BR, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,], # B14
[W_BI, 0.0, W_BR, W_BR, 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,], # B23
[W_BI, 0.0, W_BR, 0.0, W_BR, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,], # B24
[W_BI, 0.0, 0.0, W_BR, W_BR, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,], # B34
[ 0.0, 0.0, 0.0, 0.0, 0.0, W_AB, 0.0, 0.0, 0.0, 0.0, 0.0, W_AA, 0.0, 0.0, 0.0, 0.0, 0.0,], # A12
[ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, W_AB, 0.0, 0.0, 0.0, 0.0, 0.0, W_AA, 0.0, 0.0, 0.0, 0.0,], # A13
[ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, W_AB, 0.0, 0.0, 0.0, 0.0, 0.0, W_AA, 0.0, 0.0, 0.0,], # A14
[ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, W_AB, 0.0, 0.0, 0.0, 0.0, 0.0, W_AA, 0.0, 0.0,], # A23
[ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, W_AB, 0.0, 0.0, 0.0, 0.0, 0.0, W_AA, 0.0,], # A24
[ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, W_AB, 0.0, 0.0, 0.0, 0.0, 0.0, W_AA,], # A34
# I R1 R2 R3 R4 B12 B13 B14 B23 B24 B34 A12 A13 A14 A23 A24 A34
])
# build network
ntwk = network.RateBasedModel(nodes, weights)
ntwk.store_voltages = True
ntwk.noise_level = NOISE_LEVEL
# set up drive
drives = np.zeros((DURATION, len(nodes)), dtype=float)
drives[DRIVE_11_START:DRIVE_11_END, 1] = DRIVE_11_AMP
drives[DRIVE_12_START:DRIVE_12_END, 1] = DRIVE_12_AMP
drives[DRIVE_13_START:DRIVE_13_END, 1] = DRIVE_13_AMP
drives[DRIVE_21_START:DRIVE_21_END, 2] = DRIVE_21_AMP
# run network
np.random.seed(SEED)
ntwk.vs = np.array([node['v_rest'] for node in nodes])
for drive in drives:
ntwk.step(drive)
# open a figure
fig, axs = plt.subplots(4, 1, figsize=FIG_SIZE, sharex=True, tight_layout=True)
axs_twinx = [None] + [ax.twinx() for ax in axs[1:]]
# plot drive
axs[0].set_color_cycle(COLOR_CYCLE)
axs[0].plot(drives[:, 1:5], lw=2)
# plot responses of receptive field (R) units
axs[1].set_color_cycle(COLOR_CYCLE)
axs_twinx[1].set_color_cycle(COLOR_CYCLE)
axs[1].plot(np.array(ntwk.vs_history)[:, 1:5], ls='--', lw=2)
axs_twinx[1].plot(np.array(ntwk.rs_history)[:, 1:5], ls='-', lw=2)
# plot responses of buffer (B) units
axs[2].set_color_cycle(COLOR_CYCLE[4:])
axs_twinx[2].set_color_cycle(COLOR_CYCLE[4:])
axs[2].plot(np.array(ntwk.vs_history)[:, 5:11], ls='--', lw=2)
axs_twinx[2].plot(np.array(ntwk.rs_history)[:,5:11], ls='-', lw=2)
# plot response of inhibitory (I) unit
axs[2].plot(np.array(ntwk.vs_history)[:, 0], ls='--', lw=2, color='k')
axs_twinx[2].plot(np.array(ntwk.rs_history)[:, 0], ls='-', lw=2, color='k')
# plot response of association (A) units
axs[3].set_color_cycle(COLOR_CYCLE[4:])
axs_twinx[3].set_color_cycle(COLOR_CYCLE[4:])
axs[3].plot(np.array(ntwk.vs_history)[:, 11:], ls='--', lw=2)
axs_twinx[3].plot(np.array(ntwk.rs_history)[:, 11:], ls='-', lw=2)
# set axis limits
for ax in axs_twinx[1:]:
ax.set_ylim(0, 1)
# label things
axs[0].set_title('drive')
axs[0].set_ylabel('strength')
axs[1].set_title('receptive field units')
axs[1].set_ylabel('voltage')
axs_twinx[1].set_ylabel('firing rate')
axs[2].set_title('buffer units')
axs[2].set_ylabel('voltage')
axs_twinx[2].set_ylabel('firing rate')
axs[3].set_title('association units')
axs[3].set_ylabel('voltage')
axs_twinx[3].set_ylabel('firing rate')
axs[3].set_xlabel('time')
for ax in list(axs) + axs_twinx[1:]:
axis_tools.set_fontsize(ax, FONT_SIZE)
def main(config):
# unpack params from config
SEED = config['SEED']
TAU = config['TAU']
TAU_M = config['TAU_M']
TAU_C = config['TAU_C']
V_REST = config['V_REST']
V_REST_C = config['V_REST_C']
V_TH = config['V_TH']
STEEPNESS = config['STEEPNESS']
W_IF = config['W_IF']
W_FS = config['W_FS']
W_FI = config['W_FI']
W_FF = config['W_FF']
W_FC = config['W_FC']
W_MF = config['W_MF']
W_MM = config['W_MM']
W_CF = config['W_CF']
W_CM = config['W_CM']
N_UNITS = config['N_UNITS']
NOISE_LEVEL = config['NOISE_LEVEL']
S_DRIVE_AMP = config['S_DRIVE_AMP']
F0_DRIVE_AMP = config['F0_DRIVE_AMP']
F1_DRIVE_AMP = config['F1_DRIVE_AMP']
T_F0_DRIVE = config['T_F0_DRIVE']
D_F0_DRIVE = config['D_F0_DRIVE']
T_F1_DRIVE = config['T_F1_DRIVE']
D_F1_DRIVE = config['D_F1_DRIVE']
T2_F1_DRIVE = config['T2_F1_DRIVE']
D2_F1_DRIVE = config['D2_F1_DRIVE']
T2_F0_DRIVE = config['T2_F0_DRIVE']
D2_F0_DRIVE = config['D2_F0_DRIVE']
T_S_DRIVE = config['T_S_DRIVE']
DURATION = config['DURATION']
FONT_SIZE = config['FONT_SIZE']
COLOR_CYCLE = config['COLOR_CYCLE']
# generate network nodes and weights using helper function
nodes, weights = network_param_gen.wta_memory_combo(
n_units=N_UNITS,
tau=TAU, tau_m=TAU_M, tau_c=TAU_C, v_rest=V_REST, v_rest_c=V_REST_C, v_th=V_TH, steepness=STEEPNESS,
w_if=W_IF, w_fs=W_FS, w_fi=W_FI, w_ff=W_FF, w_fc=W_FC, w_mf=W_MF, w_mm=W_MM, w_cf=W_CF, w_cm=W_CM,
)
# the order of the neurons in this network is:
# s, i, f, f, ..., f, f, m, c, c, m, c, c, ..., m, c, c, m, c, c
# there are N_UNITS f neurons, and N_UNITS*(N_UNITS-1)/2 m, c, c groups
ntwk = network.RateBasedModel(nodes, weights)
ntwk.noise_level = NOISE_LEVEL
ntwk.store_voltages = True
# setup network drive
s_drive = np.zeros((DURATION, 1), dtype=float)
s_drive[T_S_DRIVE:, 0] = S_DRIVE_AMP
f_drive = np.zeros((DURATION, N_UNITS), dtype=float)
f_drive[T_F0_DRIVE:T_F0_DRIVE+D_F0_DRIVE, 0] = F0_DRIVE_AMP
f_drive[T_F1_DRIVE:T_F1_DRIVE+D_F1_DRIVE, 1] = F1_DRIVE_AMP
f_drive[T2_F1_DRIVE:T2_F1_DRIVE+D2_F1_DRIVE, 1] = F1_DRIVE_AMP
f_drive[T2_F0_DRIVE:T2_F0_DRIVE+D2_F0_DRIVE, 0] = F0_DRIVE_AMP
i_drive = np.zeros((DURATION, 1), dtype=float)
mcc_drive = np.zeros((DURATION, 3*N_UNITS*(N_UNITS-1)/2), dtype=float)
drives = np.concatenate([s_drive, i_drive, f_drive, mcc_drive], axis=1)
# run simulation
np.random.seed(SEED)
ntwk.vs = np.array([n['v_rest'] for n in nodes])
for drive in drives:
ntwk.step(drive)
# do some things before making figures
rs = np.array(ntwk.rs_history)
vs = np.array(ntwk.vs_history)
f_idxs = np.arange(2, 2 + N_UNITS, dtype=int)
m_idxs = np.arange(2 + N_UNITS, 2 + N_UNITS + 3 * N_UNITS * (N_UNITS - 1) / 2, 3, dtype=int)
fig, axs = plt.subplots(4, 1, figsize=(15, 12), sharex=True, tight_layout=True)
axs[3].twin = axs[3].twinx()
for ax in np.concatenate([axs.flatten(), [axs[3].twin]]):
ax.set_color_cycle(COLOR_CYCLE)
axs[0].plot(rs[:, f_idxs], lw=2)
axs[1].plot(drives[:, np.concatenate([f_idxs, [0]])], lw=2)
axs[2].plot(vs[:, m_idxs], lw=2)
axs[3].plot(rs[:, range(m_idxs[0] + 1, m_idxs[0] + 3)], lw=2, ls='-')
axs[3].twin.plot(vs[:, range(m_idxs[0] + 1, m_idxs[0] + 3)], lw=2, ls='--')
axs[0].set_title('Fast units')
axs[0].set_ylabel('Firing rate')
axs[1].set_title('Drive')
axs[1].set_ylabel('Drive')
axs[2].set_title('Memory units')
axs[2].set_ylabel('Voltage')
axs[3].set_title('Conduit units')
axs[3].set_ylabel('Firing rate')
axs[3].twin.set_ylabel('Voltage')
axs[3].set_xlabel('t')
for ax in np.concatenate([axs.flatten(), [axs[3].twin]]):
axis_tools.set_fontsize(ax, FONT_SIZE)
def novel_pattern_replay(CONFIG):
"""
Show how a network that has weak connections in addition to its strong connections can learn to replay novel patterns that would not normally arise spontaneously.
"""
SEED = CONFIG['SEED']
LOAD_FILE_NAME = CONFIG['LOAD_FILE_NAME']
W_WEAK = CONFIG['W_WEAK']
GAIN = CONFIG['GAIN']
REFRACTORY_STRENGTH = CONFIG['REFRACTORY_STRENGTH']
LINGERING_INPUT_VALUE = CONFIG['LINGERING_INPUT_VALUE']
LINGERING_INPUT_TIMESCALE = CONFIG['LINGERING_INPUT_TIMESCALE']
STRONG_DRIVE_AMPLITUDE = CONFIG['STRONG_DRIVE_AMPLITUDE']
WEAK_DRIVE_AMPLITUDE = CONFIG['WEAK_DRIVE_AMPLITUDE']
TRIAL_LENGTH_TRIGGERED_REPLAY = CONFIG['TRIAL_LENGTH_TRIGGERED_REPLAY']
RUN_LENGTH = CONFIG['RUN_LENGTH']
FIG_SIZE_0 = CONFIG['FIG_SIZE_0']
FIG_SIZE_1 = CONFIG['FIG_SIZE_1']
FONT_SIZE = CONFIG['FONT_SIZE']
np.random.seed(SEED)
fig = plt.figure(figsize=FIG_SIZE_0, tight_layout=True)
axs = []
for row_ctr in range(3):
axs.append([fig.add_subplot(4, 3, 3*row_ctr + col_ctr) for col_ctr in range(1, 4)])
axs = list(axs)
axs.append(fig.add_subplot(4, 1, 4))
# load old network
ntwk_old = np.load(LOAD_FILE_NAME)[0]
# demonstrate how one cannot use intrinsic plasticity to learn sequence that is made of disjoint paths
path = list(ntwk_old.node_0_path_tree[0][:])
path[2:] = ntwk_old.node_1_path_tree[0][2:]
drives = np.zeros((TRIAL_LENGTH_TRIGGERED_REPLAY, ntwk_old.w.shape[0]), dtype=float)
for ctr, node in enumerate(path):
drives[ctr, node] = STRONG_DRIVE_AMPLITUDE
drives[len(path), path[0]] = STRONG_DRIVE_AMPLITUDE
for ctr, ax in enumerate(axs[0]):
ntwk = deepcopy(ntwk_old)
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(ax, spikes, drives)
ax.set_xlim(-1, len(drives))
ax.set_ylim(-1, 20)
ax.set_xlabel('time step')
ax.set_ylabel('active \n ensemble')
ax.set_title('Strongly driving nonexisting path (trial {})'.format(ctr + 1))
w = ntwk_old.w.copy()
# add weak connection to element 2 of node_1 path tree from element 1 of node_0 path tree
w[ntwk_old.node_1_path_tree[0][2], ntwk_old.node_0_path_tree[0][1]] = W_WEAK
# make new base network
ntwk_base = network.RecurrentSoftMaxLingeringModel(
w, GAIN, REFRACTORY_STRENGTH, LINGERING_INPUT_VALUE, LINGERING_INPUT_TIMESCALE
)
ntwk_base.node_0 = ntwk_old.node_0
ntwk_base.node_0 = ntwk_old.node_1
ntwk_base.node_0_path_tree = ntwk_old.node_0_path_tree
ntwk_base.node_1_path_tree = ntwk_old.node_1_path_tree
# demonstrate how weak connections allow linking of paths into short term memory
path = list(ntwk_base.node_0_path_tree[0][:])
path[2:] = ntwk_base.node_1_path_tree[0][2:]
drives = np.zeros((TRIAL_LENGTH_TRIGGERED_REPLAY, ntwk_base.w.shape[0]), dtype=float)
for ctr, node in enumerate(path):
drives[ctr, node] = STRONG_DRIVE_AMPLITUDE
drives[len(path), path[0]] = STRONG_DRIVE_AMPLITUDE
for ctr, ax in enumerate(axs[1]):
ntwk = deepcopy(ntwk_base)
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(ax, spikes, drives)
ax.set_xlim(-1, len(drives))
ax.set_ylim(-1, 20)
ax.set_xlabel('time step')
ax.set_ylabel('active \n ensemble')
ax.set_title('Activity from forced initial condition after \n driving path with weak connection (trial {})'.format(ctr + 1))
# demonstrate how weak connections do not substantially affect path probabilities
path = list(ntwk_base.node_0_path_tree[0][:])
path[2:] = ntwk_base.node_1_path_tree[0][2:]
drives = np.zeros((TRIAL_LENGTH_TRIGGERED_REPLAY, ntwk_old.w.shape[0]), dtype=float)
drives[0, path[0]] = STRONG_DRIVE_AMPLITUDE
for ctr, ax in enumerate(axs[2]):
ntwk = deepcopy(ntwk_base)
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(ax, spikes, drives)
ax.set_xlim(-1, len(drives))
ax.set_ylim(-1, 20)
ax.set_xlabel('time step')
ax.set_ylabel('active ensemble')
ax.set_title('Free activity after only forcing \n initial condition (trial {})'.format(ctr + 1))
path = list(ntwk_base.node_0_path_tree[0][:])
path[2:] = ntwk_base.node_1_path_tree[0][2:]
drives = np.zeros((RUN_LENGTH, ntwk_base.w.shape[0]), dtype=float)
for ctr, node in enumerate(path):
drives[ctr, node] = STRONG_DRIVE_AMPLITUDE
ntwk = deepcopy(ntwk_base)
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[3], spikes, drives)
axs[3].set_xlim(-1, len(drives))
axs[3].set_ylim(-1, 40)
axs[3].set_xlabel('time step')
axs[3].set_ylabel('active \n ensemble')
axs[3].set_title('Free activity after driving path with weak connection')
for ax_row in axs[:-1]:
for ax in ax_row:
axis_tools.set_fontsize(ax, FONT_SIZE)
axis_tools.set_fontsize(axs[-1], FONT_SIZE)
# now demonstrate how pattern-matching computation changes with respect to short-term memory
fig, axs = plt.subplots(1, 2, figsize=FIG_SIZE_1, tight_layout=True)
path = list(ntwk_base.node_0_path_tree[1][:])
path[0] = 22
path[2] = 17
path[3] = ntwk_base.node_1_path_tree[0][3]
drives_new = np.zeros((len(path), ntwk_base.w.shape[0]), dtype=float)
drives_new[0, path[0]] = STRONG_DRIVE_AMPLITUDE
for ctr, node in enumerate(path[1:]):
drives_new[ctr + 1, node] = WEAK_DRIVE_AMPLITUDE
# drive a network with just the new drive to see how it completes the pattern
ntwk = deepcopy(ntwk_base)
ntwk.store_voltages = True
for drive in drives_new:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[0], spikes, drives_new)
axs[0].set_xlim(-1, 8)
axs[0].set_ylim(-1, ntwk_base.w.shape[0])
axs[0].set_xlabel('time step')
axs[0].set_ylabel('active ensemble')
axs[0].set_title('Weakly driving nonexistent path')
drives = np.concatenate([drives[:4, :], drives_new])
ntwk = deepcopy(ntwk_base)
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[1], spikes, drives)
axs[1].set_xlim(-1, 8)
axs[1].set_ylim(-1, ntwk_base.w.shape[0])
axs[1].set_xlabel('time step')
axs[1].set_ylabel('active ensemble')
axs[1].set_title('Weakly driving nonexistent path after \n strongly driving path with weak connection')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
def basic_replay_stats(CONFIG):
"""
Plot some statistics of this network's behavior.
"""
SEED = CONFIG['SEED']
LOAD_FILE_NAME = CONFIG['LOAD_FILE_NAME']
GAIN_HIGH = CONFIG['GAIN_HIGH']
GAIN_LOW = CONFIG['GAIN_LOW']
LINGERING_INPUT_VALUE = CONFIG['LINGERING_INPUT_VALUE']
LINGERING_INPUT_TIMESCALE = CONFIG['LINGERING_INPUT_TIMESCALE']
STRONG_DRIVE_AMPLITUDE = CONFIG['STRONG_DRIVE_AMPLITUDE']
T_SPONTANEOUS = CONFIG['T_SPONTANEOUS']
SPONTANEOUS_REPEATS = CONFIG['SPONTANEOUS_REPEATS']
FIG_SIZE = CONFIG['FIG_SIZE']
FONT_SIZE = CONFIG['FONT_SIZE']
np.random.seed(SEED)
fig, axs = plt.subplots(1, 3, figsize=FIG_SIZE, tight_layout=True)
ntwk_base = np.load(LOAD_FILE_NAME)[0]
n_nodes = ntwk_base.w.shape[0]
# show that adding in lingering activity increases probability of replay
driven_path = ntwk_base.node_0_path_tree[0]
p_replay_no_lingering = 1
for node_prev, node_next in zip(driven_path[:-1], driven_path[1:]):
intrinsic = ntwk_base.w[:, node_prev]
refractory = np.zeros((n_nodes,), dtype=float)
refractory[node_prev] = ntwk_base.refractory_strength
inputs = intrinsic + refractory
prob = np.exp(ntwk_base.gain * inputs)
prob /= prob.sum()
p_replay_no_lingering *= prob[node_next]
lingering_inputs = np.zeros((n_nodes,), dtype=float)
lingering_inputs[np.array(driven_path)] = LINGERING_INPUT_VALUE
p_replay_with_lingering = 1
for node_prev, node_next in zip(driven_path[:-1], driven_path[1:]):
intrinsic = ntwk_base.w[:, node_prev]
refractory = np.zeros((n_nodes,), dtype=float)
refractory[node_prev] = ntwk_base.refractory_strength
inputs = intrinsic + refractory + lingering_inputs
prob = np.exp(ntwk_base.gain * inputs)
prob /= prob.sum()
p_replay_with_lingering *= prob[node_next]
axs[0].bar([0, 1], [p_replay_no_lingering, p_replay_with_lingering], align='center')
axs[0].set_xticks([0, 1])
axs[0].set_xticklabels(['Without \n nonassociative \n priming', 'With \n nonassociative \n priming'])
axs[0].set_ylabel('sequence replay probability')
for ax, gain in zip(axs[1:], [GAIN_LOW, GAIN_HIGH]):
# show expected number of spontaneous replays with and without nonassociative priming
ntwk_no_nap = deepcopy(ntwk_base)
ntwk_no_nap.gain = gain
ntwk_with_nap = deepcopy(ntwk_base)
ntwk_with_nap.gain = gain
ntwk_with_nap.lingering_input_value = LINGERING_INPUT_VALUE
ntwk_with_nap.lingering_input_timescale = LINGERING_INPUT_TIMESCALE
drives = np.zeros((T_SPONTANEOUS, n_nodes), dtype=float)
for t, node in enumerate(driven_path):
drives[t, node] = STRONG_DRIVE_AMPLITUDE
past_occurrences_of_driven_seq_no_nap = []
for _ in range(SPONTANEOUS_REPEATS):
ntwk = deepcopy(ntwk_no_nap)
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
activation_seq = np.array(ntwk.rs_history).nonzero()[1]
past_occurrences_of_driven_seq_no_nap.append(
metrics.get_number_of_past_occurrences_of_specific_sequence(activation_seq, driven_path)
)
past_occurrences_of_driven_seq_no_nap = np.array(past_occurrences_of_driven_seq_no_nap)
past_occurrences_of_driven_seq_no_nap_mean = np.mean(past_occurrences_of_driven_seq_no_nap, axis=0)
past_occurrences_of_driven_seq_no_nap_sem = stats.sem(past_occurrences_of_driven_seq_no_nap, axis=0)
past_occurrences_of_driven_seq_with_nap = []
for _ in range(SPONTANEOUS_REPEATS):
ntwk = deepcopy(ntwk_with_nap)
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
activation_seq = np.array(ntwk.rs_history).nonzero()[1]
past_occurrences_of_driven_seq_with_nap.append(
metrics.get_number_of_past_occurrences_of_specific_sequence(activation_seq, driven_path)
)
past_occurrences_of_driven_seq_with_nap = np.array(past_occurrences_of_driven_seq_with_nap)
past_occurrences_of_driven_seq_with_nap_mean = np.mean(past_occurrences_of_driven_seq_with_nap, axis=0)
past_occurrences_of_driven_seq_with_nap_sem = stats.sem(past_occurrences_of_driven_seq_with_nap, axis=0)
ts = np.arange(T_SPONTANEOUS)
ax.plot(ts, past_occurrences_of_driven_seq_no_nap_mean, color='b', lw=2)
ax.fill_between(
ts,
past_occurrences_of_driven_seq_no_nap_mean - past_occurrences_of_driven_seq_no_nap_sem,
past_occurrences_of_driven_seq_no_nap_mean + past_occurrences_of_driven_seq_no_nap_sem,
color='b',
alpha=0.3,
)
ax.plot(ts, past_occurrences_of_driven_seq_with_nap_mean, color='g', lw=2)
ax.fill_between(
ts,
past_occurrences_of_driven_seq_with_nap_mean - past_occurrences_of_driven_seq_with_nap_sem,
past_occurrences_of_driven_seq_with_nap_mean + past_occurrences_of_driven_seq_with_nap_sem,
color='g',
alpha=0.3,
)
ax.set_xlabel('time step')
ax.set_ylabel('past occurrences')
ax.set_title('gain = {}'.format(gain))
ax.legend(['Without nonasso-\n ciative priming', 'With nonasso- \n ciative priming'], loc='best')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
def basic_replay_ex(CONFIG):
"""
Run a simulation demonstrating the basic capability of a network with nonassociative priming to
demonstrated replay, both triggered and spontaneous.
"""
SEED = CONFIG['SEED']
LOAD_FILE_NAME = CONFIG['LOAD_FILE_NAME']
GAIN_HIGH = CONFIG['GAIN_HIGH']
GAIN_LOW = CONFIG['GAIN_LOW']
LINGERING_INPUT_VALUE = CONFIG['LINGERING_INPUT_VALUE']
LINGERING_INPUT_TIMESCALE = CONFIG['LINGERING_INPUT_TIMESCALE']
STRONG_DRIVE_AMPLITUDE = CONFIG['STRONG_DRIVE_AMPLITUDE']
WEAK_DRIVE_AMPLITUDE = CONFIG['WEAK_DRIVE_AMPLITUDE']
TRIAL_LENGTH_TRIGGERED_REPLAY = CONFIG['TRIAL_LENGTH_TRIGGERED_REPLAY']
RUN_LENGTH = CONFIG['RUN_LENGTH']
FIG_SIZE = CONFIG['FIG_SIZE']
FONT_SIZE = CONFIG['FONT_SIZE']
np.random.seed(SEED)
ntwk_base = np.load(LOAD_FILE_NAME)[0]
ntwk_base.lingering_input_value = LINGERING_INPUT_VALUE
ntwk_base.lingering_input_timescale = LINGERING_INPUT_TIMESCALE
fig = plt.figure(figsize=FIG_SIZE, tight_layout=True)
axs = []
axs.append(fig.add_subplot(6, 2, 1))
axs.append(fig.add_subplot(6, 2, 2))
axs.append(fig.add_subplot(6, 2, 3))
axs.append(fig.add_subplot(6, 2, 4))
axs.append(fig.add_subplot(6, 1, 3))
axs.append(fig.add_subplot(6, 1, 4))
axs.append(fig.add_subplot(6, 1, 5))
axs.append(fig.add_subplot(6, 1, 6))
# play sequences aligned to the network's intrinsic path structure
path_00 = ntwk_base.node_0_path_tree[0]
path_10 = ntwk_base.node_1_path_tree[0]
# drive network for first trial: path_00
drives = np.zeros((TRIAL_LENGTH_TRIGGERED_REPLAY, ntwk_base.w.shape[1]), dtype=float)
drives[0, path_00[0]] = STRONG_DRIVE_AMPLITUDE
for t_ctr, node in enumerate(path_00[1:]):
drives[t_ctr + 1, node] = WEAK_DRIVE_AMPLITUDE
drives[len(path_00), path_00[0]] = STRONG_DRIVE_AMPLITUDE
ntwk = deepcopy(ntwk_base)
ntwk.gain = GAIN_HIGH
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[0], spikes, drives)
axs[0].set_xlim(-1, len(drives))
axs[0].set_ylim(-1, 20)
axs[0].set_xlabel('time step')
axs[0].set_ylabel('active ensemble')
axs[0].set_title('Aligning external drive with \n strongly connected paths')
# drive network for first trial: path_10
drives = np.zeros((TRIAL_LENGTH_TRIGGERED_REPLAY, ntwk_base.w.shape[1]), dtype=float)
drives[0, path_10[0]] = STRONG_DRIVE_AMPLITUDE
for t_ctr, node in enumerate(path_10[1:]):
drives[t_ctr + 1, node] = WEAK_DRIVE_AMPLITUDE
drives[len(path_10), path_10[0]] = STRONG_DRIVE_AMPLITUDE
ntwk = deepcopy(ntwk_base)
ntwk.gain = GAIN_HIGH
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[1], spikes, drives)
axs[1].set_xlim(-1, len(drives))
axs[1].set_ylim(-1, 20)
axs[1].set_xlabel('time step')
axs[1].set_ylabel('active ensemble')
axs[1].set_title('Aligning external drive with \n strongly connected paths')
# drive network for third trial: all path_00 except for element 2
path = list(path_00[:])
path[2] = path_10[2]
drives = np.zeros((TRIAL_LENGTH_TRIGGERED_REPLAY, ntwk_base.w.shape[1]), dtype=float)
drives[0, path[0]] = STRONG_DRIVE_AMPLITUDE
for t_ctr, node in enumerate(path[1:]):
drives[t_ctr + 1, node] = WEAK_DRIVE_AMPLITUDE
drives[len(path), path[0]] = STRONG_DRIVE_AMPLITUDE
ntwk = deepcopy(ntwk_base)
ntwk.gain = GAIN_HIGH
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[2], spikes, drives)
axs[2].set_xlim(-1, len(drives))
axs[2].set_ylim(-1, 20)
axs[2].set_xlabel('time step')
axs[2].set_ylabel('active ensemble')
axs[2].set_title('Aligning external drive with \n nonexisting path')
# drive network for fourth trial: all path_10 except for element 2
path = list(path_10[:])
path[2] = path_00[2]
drives = np.zeros((TRIAL_LENGTH_TRIGGERED_REPLAY, ntwk_base.w.shape[1]), dtype=float)
drives[0, path[0]] = STRONG_DRIVE_AMPLITUDE
for t_ctr, node in enumerate(path[1:]):
drives[t_ctr + 1, node] = WEAK_DRIVE_AMPLITUDE
drives[len(path), path[0]] = STRONG_DRIVE_AMPLITUDE
ntwk = deepcopy(ntwk_base)
ntwk.gain = GAIN_HIGH
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[3], spikes, drives)
axs[3].set_xlim(-1, len(drives))
axs[3].set_ylim(-1, 20)
axs[3].set_xlabel('time step')
axs[3].set_ylabel('active ensemble')
axs[3].set_title('Aligning external drive with \n nonexisting path')
# play sequence and then let network run spontaneously for a while
drives = np.zeros((RUN_LENGTH, ntwk_base.w.shape[1]), dtype=float)
for t_ctr, node in enumerate(path_00):
drives[t_ctr, node] = STRONG_DRIVE_AMPLITUDE
ntwk = deepcopy(ntwk_base)
ntwk.gain = GAIN_HIGH
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[4], spikes, drives)
axs[4].set_xlim(-1, len(drives))
axs[4].set_ylim(-1, ntwk_base.w.shape[1])
axs[4].set_xlabel('time step')
axs[4].set_ylabel('active ensemble')
axs[4].set_title('Letting network run freely after driving strongly connected path (high gain)')
# play sequence and then let network run spontaneously for a while, now with lower gain
drives = np.zeros((RUN_LENGTH, ntwk_base.w.shape[1]), dtype=float)
for t_ctr, node in enumerate(path_00):
drives[t_ctr, node] = STRONG_DRIVE_AMPLITUDE
ntwk = deepcopy(ntwk_base)
ntwk.gain = GAIN_LOW
ntwk.store_voltages = True
for drive in drives:
ntwk.step(drive)
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[5], spikes, drives)
axs[5].set_xlim(-1, len(drives))
axs[5].set_ylim(-1, ntwk_base.w.shape[1])
axs[5].set_xlabel('time step')
axs[5].set_ylabel('active ensemble')
axs[5].set_title('Letting network run freely after driving strongly connected path (low gain)')
# let network run spontaneously for a while with no initial drive
ntwk = deepcopy(ntwk_base)
ntwk.gain = GAIN_HIGH
ntwk.store_voltages = True
for _ in range(RUN_LENGTH):
ntwk.step()
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[6], spikes, drives=None)
axs[6].set_xlim(-1, len(drives))
axs[6].set_ylim(-1, ntwk_base.w.shape[1])
axs[6].set_xlabel('time step')
axs[6].set_ylabel('active ensemble')
axs[6].set_title('Letting network run freely with no drive (high gain)')
# let network run spontaneously for a while with no initial drive, now with lower gain
ntwk = deepcopy(ntwk_base)
ntwk.gain = GAIN_LOW
ntwk.store_voltages = True
for _ in range(RUN_LENGTH):
ntwk.step()
spikes = np.array(ntwk.rs_history)
fancy_raster.by_row_circles(axs[7], spikes, drives=None)
axs[7].set_xlim(-1, len(drives))
axs[7].set_ylim(-1, ntwk_base.w.shape[1])
axs[7].set_xlabel('time step')
axs[7].set_ylabel('active ensemble')
axs[7].set_title('Letting network run freely with no drive (low gain)')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
def main(config):
# get params from config
SEED = config['SEED']
DURATION = config['DURATION']
TAU = config['TAU']
V_REST = config['V_REST']
THRESHOLD = config['THRESHOLD']
STEEPNESS = config['STEEPNESS']
NOISE_LEVEL = config['NOISE_LEVEL']
W_FS = config['W_FS']
W_FF = config['W_FF']
W_FI = config['W_FI']
W_XFI = config['W_XFI']
W_IF = config['W_IF']
SWITCH_DRIVE = config['SWITCH_DRIVE']
FONT_SIZE = config['FONT_SIZE']
COLOR_CYCLE = config['COLOR_CYCLE']
## run simulation with single WTA unit
# setup network parameters; order: (switch, fast, fast_inh)
w_single = np.array([
[0, 0, 0], # to switch
[W_FS, W_FF, W_FI], # to fast
[0, W_IF, 0], # to fast_inh
])
nodes_single = 3 * [{'tau': TAU, 'v_rest': V_REST, 'threshold': THRESHOLD, 'steepness': STEEPNESS}]
# build network
ntwk = network.RateBasedModel(nodes_single, w_single)
ntwk.noise_level = NOISE_LEVEL
ntwk.store_voltages = True
# set network drive
drive = np.array([SWITCH_DRIVE, 0, 0])
# run network
np.random.seed(SEED)
ntwk.vs = np.array([V_REST, V_REST, V_REST])
for t in range(DURATION):
ntwk.step(drive)
# make figure
fig, axs = plt.subplots(2, 1, figsize=(15, 5), sharex=True, tight_layout=True)
for ax in axs:
ax.set_color_cycle(COLOR_CYCLE)
axs[0].plot(ntwk.vs_history, lw=2)
axs[1].plot(ntwk.rs_history, lw=2)
axs[1].set_ylim(-.1, 1.1)
axs[1].set_yticks(np.linspace(0, 1, 5, endpoint=True))
axs[0].set_ylabel('Voltage')
axs[1].set_xlabel('t')
axs[1].set_ylabel('Firing rate')
axs[0].set_title('Single WTA Unit')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
## run simulation with three non-interacting WTA units
# setup network parameters; order: (switch, fast, fast_inh, fast, fast_inh, fast, fast_inh)
w_multi = np.array([
[0, 0, 0, 0, 0, 0, 0], # to switch
[W_FS, W_FF, W_FI, 0, 0, 0, 0], # to fast1
[0, W_IF, 0, 0, 0, 0, 0], # to fast_inh1
[W_FS, 0, 0, W_FF, W_FI, 0, 0], # to fast2
[0, 0, 0, W_IF, 0, 0, 0], # to fast_inh2
[W_FS, 0, 0, 0, 0, W_FF, W_FI], # to fast3
[0, 0, 0, 0, 0, W_IF, 0], # to fast_inh3
])
nodes_multi = 7 * [{'tau': TAU, 'v_rest': V_REST, 'threshold': THRESHOLD, 'steepness': STEEPNESS}]
# build network
ntwk = network.RateBasedModel(nodes_multi, w_multi)
ntwk.noise_level = NOISE_LEVEL
ntwk.store_voltages = True
# set network drive
drive = np.array([SWITCH_DRIVE, 0, 0, 0, 0, 0, 0])
# run network
np.random.seed(SEED)
ntwk.vs = np.array([V_REST, V_REST, V_REST, V_REST, V_REST, V_REST, V_REST])
for t in range(DURATION):
ntwk.step(drive)
# make figure
fig, axs = plt.subplots(2, 1, figsize=(15, 5), sharex=True, tight_layout=True)
for ax in axs:
ax.set_color_cycle(COLOR_CYCLE)
axs[0].plot(np.array(ntwk.vs_history)[:, [1, 3, 5]], lw=2)
axs[1].plot(np.array(ntwk.rs_history)[:, [1, 3, 5]], lw=2)
axs[1].set_ylim(-.1, 1.1)
axs[1].set_yticks(np.linspace(0, 1, 5, endpoint=True))
axs[0].set_ylabel('Voltage')
axs[1].set_xlabel('t')
axs[1].set_ylabel('Firing rate')
axs[0].set_title('Three Independent WTA Units')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
## run simulation with three interacting WTA units
# setup network parameters; order: (switch, fast1, fast_inh1, fast2, fast_inh2, fast3, fast_inh3)
w_multi = np.array([
[0, 0, 0, 0, 0, 0, 0], # to switch
[W_FS, W_FF, W_FI, 0, W_XFI, 0, W_XFI], # to fast1
[0, W_IF, 0, 0, 0, 0, 0], # to fast_inh1
[W_FS, 0, W_XFI, W_FF, W_FI, 0, W_XFI], # to fast2
[0, 0, 0, W_IF, 0, 0, 0], # to fast_inh2
[W_FS, 0, W_XFI, 0, W_XFI, W_FF, W_FI], # to fast3
[0, 0, 0, 0, 0, W_IF, 0], # to fast_inh3
])
nodes_multi = 7 * [{'tau': TAU, 'v_rest': V_REST, 'threshold': THRESHOLD, 'steepness': STEEPNESS}]
# build network
ntwk = network.RateBasedModel(nodes_multi, w_multi)
ntwk.noise_level = NOISE_LEVEL
ntwk.store_voltages = True
# set network drive
drive = np.array([SWITCH_DRIVE, 0, 0, 0, 0, 0, 0])
# run network
np.random.seed(SEED)
ntwk.vs = np.array([V_REST, V_REST, V_REST, V_REST, V_REST, V_REST, V_REST])
for t in range(DURATION):
ntwk.step(drive)
# make figure
fig, axs = plt.subplots(2, 1, figsize=(15, 5), sharex=True, tight_layout=True)
for ax in axs:
ax.set_color_cycle(COLOR_CYCLE)
axs[0].plot(np.array(ntwk.vs_history)[:, [1, 3, 5]], lw=2)
axs[1].plot(np.array(ntwk.rs_history)[:, [1, 3, 5]], lw=2)
axs[1].set_ylim(-.1, 1.1)
axs[1].set_yticks(np.linspace(0, 1, 5, endpoint=True))
axs[0].set_ylabel('Voltage')
axs[1].set_xlabel('t')
axs[1].set_ylabel('Firing rate')
axs[0].set_title('Three Interacting WTA Units')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
## run simulation with three fast units all of whose interactions occur through one inhibitory unit
# setup network parameters; order: (switch, inh, fast1, fast2, fast3)
w_multi = np.array([
[0, 0, 0, 0, 0], # to switch
[0, 0, W_IF, W_IF, W_IF], # to inh
[W_FS, W_FI, W_FF, 0, 0], # to fast1
[W_FS, W_FI, 0, W_FF, 0], # to fast2
[W_FS, W_FI, 0, 0, W_FF], # to fast3
])
nodes_multi = 5 * [{'tau': TAU, 'v_rest': V_REST, 'threshold': THRESHOLD, 'steepness': STEEPNESS}]
# build network
ntwk = network.RateBasedModel(nodes_multi, w_multi)
ntwk.noise_level = NOISE_LEVEL
ntwk.store_voltages = True
# set network drive
drive = np.array([SWITCH_DRIVE, 0, 0, 0, 0])
# run network
np.random.seed(SEED)
ntwk.vs = np.array([V_REST, V_REST, V_REST, V_REST, V_REST])
for t in range(DURATION):
ntwk.step(drive)
# make figure
fig, axs = plt.subplots(2, 1, figsize=(15, 5), sharex=True, tight_layout=True)
for ax in axs:
ax.set_color_cycle(COLOR_CYCLE)
axs[0].plot(np.array(ntwk.vs_history)[:, 2:], lw=2)
axs[1].plot(np.array(ntwk.rs_history)[:, 2:], lw=2)
axs[1].set_ylim(-.1, 1.1)
axs[1].set_yticks(np.linspace(0, 1, 5, endpoint=True))
axs[0].set_ylabel('Voltage')
axs[1].set_xlabel('t')
axs[1].set_ylabel('Firing rate')
axs[0].set_title('Three Exc Units Sharing One Inh Unit')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
## run simulation with many fast units all of whose interactions occur through one inhibitory unit
# setup network parameters; order: (switch, inh, fast1, fast2, fast3)
w_multi = np.array([
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0], # to switch
[0, 0, W_IF, W_IF, W_IF, W_IF, W_IF, W_IF, W_IF, W_IF], # to inh
[W_FS, W_FI, W_FF, 0, 0, 0, 0, 0, 0, 0], # to fast1
[W_FS, W_FI, 0, W_FF, 0, 0, 0, 0, 0, 0], # to fast2
[W_FS, W_FI, 0, 0, W_FF, 0, 0, 0, 0, 0], # to fast3
[W_FS, W_FI, 0, 0, 0, W_FF, 0, 0, 0, 0], # to fast3
[W_FS, W_FI, 0, 0, 0, 0, W_FF, 0, 0, 0], # to fast3
[W_FS, W_FI, 0, 0, 0, 0, 0, W_FF, 0, 0], # to fast3
[W_FS, W_FI, 0, 0, 0, 0, 0, 0, W_FF, 0], # to fast3
[W_FS, W_FI, 0, 0, 0, 0, 0, 0, 0, W_FF], # to fast3
])
nodes_multi = 10 * [{'tau': TAU, 'v_rest': V_REST, 'threshold': THRESHOLD, 'steepness': STEEPNESS}]
# build network
ntwk = network.RateBasedModel(nodes_multi, w_multi)
ntwk.noise_level = NOISE_LEVEL
ntwk.store_voltages = True
# set network drive
drive = np.array([SWITCH_DRIVE, 0, 0, 0, 0, 0, 0, 0, 0, 0])
# run network
np.random.seed(SEED)
ntwk.vs = np.array([V_REST, V_REST, V_REST, V_REST, V_REST, V_REST, V_REST, V_REST, V_REST, V_REST])
for t in range(DURATION):
ntwk.step(drive)
# make figure
fig, axs = plt.subplots(2, 1, figsize=(15, 5), sharex=True, tight_layout=True)
for ax in axs:
ax.set_color_cycle(COLOR_CYCLE)
axs[0].plot(np.array(ntwk.vs_history)[:, 2:], lw=2)
axs[1].plot(np.array(ntwk.rs_history)[:, 2:], lw=2)
axs[1].set_ylim(-.1, 1.1)
axs[1].set_yticks(np.linspace(0, 1, 5, endpoint=True))
axs[0].set_ylabel('Voltage')
axs[1].set_xlabel('t')
axs[1].set_ylabel('Firing rate')
axs[0].set_title('Eight Exc Units Sharing One Inh Unit')
for ax in axs:
axis_tools.set_fontsize(ax, FONT_SIZE)
