def compute_distr_ns(states, W, b_v, b_h, steps=10, T=t_sim * 1000):
nT = int(T)
distr = kl_tools.states2distr(states[:nT], N_v + N_c + N_h) + 1
distr /= distr.sum()
res_distr_cut = np.zeros(steps)
nTi = np.logspace(2, np.log10(nT), 10, base=10)
for i in range(steps):
d = kl_tools.states2distr(states[:nTi[i]], N_v + N_c + N_h) + 1
d /= d.sum()
res_distr_cut[i] = kl_divergence_pdf(d, W, b_v, b_h) / entropy(d)
return distr, nTi, res_distr_cut
def compute_distr_ns(states, W, b_v, b_h, steps=10, T=t_sim*1000):
nT = int(T)
distr = kl_tools.states2distr(states[:nT],N_v+N_c+N_h)+1
distr/= distr.sum()
res_distr_cut = np.zeros(steps)
nTi = np.logspace(2,np.log10(nT),10,base=10)
for i in range(steps):
d = kl_tools.states2distr(states[ :nTi[i] ], N_v+N_c+N_h)+1
d /= d.sum()
res_distr_cut[i] = kl_divergence_pdf(d, W, b_v, b_h)/entropy(d)
return distr, nTi, res_distr_cut
def wrap_run(runID=0):
if isinstance(runID, int):
r, W, b_v, b_h = run_NS()
elif isinstance(runID, dict):
r, W, b_v, b_h = run_NS(runID)
states_ns = []
for rr in r:
states = states_NS(rr['Mv'], rr['Mh'], t_conv=1.0)
distr = kl_tools.states2distr(states, N_v + N_c + N_h) + 1
states_ns.append(states)
return states_ns, W, b_v, b_h
def wrap_run(runID = 0):
if isinstance(runID, int):
r, W, b_v, b_h = run_NS()
elif isinstance(runID, dict):
r, W, b_v, b_h = run_NS(runID)
states_ns = []
for rr in r:
states = states_NS(rr['Mv'], rr['Mh'], t_conv = 1.0)
distr = kl_tools.states2distr(states,N_v+N_c+N_h)+1
states_ns.append(states)
return states_ns, W, b_v, b_h