c1_abv_p = np.zeros(0)
c1_abv_1 = np.zeros(0)
c1_blo_p = np.zeros(0)
c1_blo_1 = np.zeros(0)
c2_abv_p = np.zeros(0)
c2_abv_1 = np.zeros(0)
c2_blo_p = np.zeros(0)
c2_blo_1 = np.zeros(0)
#init weights
lif.W = np.array([w0, w1])
count = np.zeros(n)
#Simulate LIF
(v_raw, h_raw, _, _, u_raw) = lif.simulate(DeltaT)
s1 = np.convolve(h_raw[0,:], exp_filter)[0:h_raw.shape[1]]
s2 = np.convolve(h_raw[1,:], exp_filter)[0:h_raw.shape[1]]
V = np.zeros((n, q))
V_mean = np.zeros((n, q))
nB = h_raw.shape[1]/DeltaT
abvthr = np.zeros(n)
blothr = np.zeros(n)
#Create a perturbed set of trains
for idx2, perturb_rate in enumerate(mvals):
U = np.zeros((n, q))
dU = np.zeros(U.shape)
c1_abv_p = np.zeros(0)
c1_abv_1 = np.zeros(0)
c1_blo_p = np.zeros(0)
c1_blo_1 = np.zeros(0)
c2_abv_p = np.zeros(0)
c2_abv_1 = np.zeros(0)
c2_blo_p = np.zeros(0)
c2_blo_1 = np.zeros(0)
#init weights
lif.W = np.array([w0, w1])
count = np.zeros(n)
#Simulate LIF
(v_raw, h_raw, _, _) = lif.simulate()
s1 = np.convolve(h_raw[0,:], exp_filter)[0:h_raw.shape[1]]
s2 = np.convolve(h_raw[1,:], exp_filter)[0:h_raw.shape[1]]
V = np.zeros((n, q))
V_mean = np.zeros((n, q))
nB = h_raw.shape[1]/DeltaT
abvthr = np.zeros(n)
blothr = np.zeros(n)
#Create a perturbed set of trains
for idx2, perturb_rate in enumerate(mvals):
U = np.zeros((n, q))
dU = np.zeros(U.shape)
N = 19
nsims = 1
wmax = 20
n = params.n
deltaT = 50
#Play with different c values
cvals = [0.01, 0.25, 0.5, 0.75, 0.99]
for c in cvals:
print("Running simulations for c = %f"%c)
outfile = './sweeps/param_w_N_%d_nsims_%d_c_%f_deltaT_counterfactual_simulations.npz'%(N, nsims, c)
params.c = c
lif.setup(params)
lif.x = 0
wvals = np.linspace(2, wmax, N)
vs = np.zeros((N, N, nsims, n, lif.T), dtype=np.float16)
hs = np.zeros((N, N, nsims, n, lif.T), dtype=np.bool)
us = np.zeros((N, N, nsims, n, lif.T), dtype=np.float16)
for i,w0 in enumerate(wvals):
for j,w1 in enumerate(wvals):
print("Running %d simulations with w0=%f, w1=%f"%(nsims, w0, w1))
lif.W = np.array([w0, w1])
for k in range(nsims):
(v, h, Cost, betas, u) = lif.simulate(deltaT)
vs[i,j,k,:] = v
hs[i,j,k,:] = h
us[i,j,k,:] = u
#Save output
np.savez(outfile, vs = vs, hs = hs, params = params, wvals = wvals\
, nsims = nsims, us = us)
se_beta_fd_c = np.zeros((len(cvals), params.n))
m_beta_rd_c_linear = np.zeros((len(cvals), params.n))
se_beta_fd_c_linear = np.zeros((len(cvals), params.n))
m_beta_rd_c_linear = np.zeros((len(cvals), params.n))
se_beta_fd_c_linear = np.zeros((len(cvals), params.n))
beta_sp_c = np.zeros((len(cvals), params.n))
cost = lambda s1, s2: (alpha1 * s1 + alpha2 * s2 - x**2)**2
for i, c in enumerate(cvals):
print("Running %d simulations with c=%s" % (nsims, c))
params.c = c
lif.setup(params)
for j in range(nsims):
(v, h, _, _, u) = lif.simulate(DeltaT)
s1 = np.convolve(h[0, :], exp_filter)[0:h.shape[1]]
s2 = np.convolve(h[1, :], exp_filter)[0:h.shape[1]]
cost_s = cost(s1, s2)
for k, p in enumerate(pvals):
#print("p = %f"%p)
beta_rd_c[i, j, k, :] = causaleffect_maxv(u, cost_s, DeltaT, p,
params)
beta_rd_c_linear[i, j, k, :] = causaleffect_maxv_linear(
u, cost_s, DeltaT, p, params)
beta_fd_c[i, j, :] = causaleffect_maxv(u, cost_s, DeltaT, 1, params)
beta_fd_c_linear[i, j, :] = causaleffect_maxv_linear(
u, cost_s, DeltaT, 1, params)
#Compute the SP cost
beta_sp_c[i, :] = causaleffect_maxv_sp(u, h, cost, DeltaT, params,
exp_filter)
#Simulate for a range of $W$ values.
N = 19
nsims = 1
wmax = 20
n = params.n
#Play with different c values
cvals = [0.01, 0.25, 0.5, 0.75, 0.99]
for c in cvals:
print("Running simulations for c = %f" % c)
outfile = './sweeps/param_w_N_%d_nsims_%d_c_%f_deltaT_simulations.npz' % (
N, nsims, c)
params.c = c
lif.setup(params)
lif.x = 0
wvals = np.linspace(2, wmax, N)
vs = np.zeros((N, N, nsims, n, lif.T), dtype=np.float16)
hs = np.zeros((N, N, nsims, n, lif.T), dtype=np.bool)
for i, w0 in enumerate(wvals):
for j, w1 in enumerate(wvals):
print("Running %d simulations with w0=%f, w1=%f" % (nsims, w0, w1))
lif.W = np.array([w0, w1])
for k in range(nsims):
(v, h, Cost, betas) = lif.simulate()
vs[i, j, k, :] = v
hs[i, j, k, :] = h
#Save output
np.savez(outfile, vs = vs, hs = hs, params = params, wvals = wvals\
, nsims = nsims)
se_beta_fd_c = np.zeros((len(cvals), params.n))
m_beta_rd_c_linear = np.zeros((len(cvals), params.n))
se_beta_fd_c_linear = np.zeros((len(cvals), params.n))
m_beta_rd_c_linear = np.zeros((len(cvals), params.n))
se_beta_fd_c_linear = np.zeros((len(cvals), params.n))
beta_sp_c = np.zeros((len(cvals), params.n))
cost = lambda s1, s2: (alpha1 * s1 + alpha2 * s2 - x**2)**2
for i, c in enumerate(cvals):
print("Running %d simulations with c=%s" % (nsims, c))
params.c = c
lif.setup(params)
for j in range(nsims):
(v, h, _, _) = lif.simulate()
s1 = np.convolve(h[0, :], exp_filter)[0:h.shape[1]]
s2 = np.convolve(h[1, :], exp_filter)[0:h.shape[1]]
cost_s = cost(s1, s2)
for k, p in enumerate(pvals):
#print("p = %f"%p)
beta_rd_c[i, j, k, :] = causaleffect_maxv(v, cost_s, DeltaT, p,
params)
beta_rd_c_linear[i, j, k, :] = causaleffect_maxv_linear(
v, cost_s, DeltaT, p, params)
beta_fd_c[i, j, :] = causaleffect_maxv(v, cost_s, DeltaT, 1, params)
beta_fd_c_linear[i, j, :] = causaleffect_maxv_linear(
v, cost_s, DeltaT, 1, params)
#Compute the SP cost
beta_sp_c[i, :] = causaleffect_maxv_sp(v, h, cost, DeltaT, params,
exp_filter)