def postVisualizer():
N_phi = 20
print 'N_phi = ', N_phi
phi_norms = linspace(1/(2.*N_phi), 1. - 1/ (2.*N_phi), N_phi)
theta = 20
base_name = 'sinusoidal_spike_train_N=1000_critical_theta=%d'%theta
T_thresh = 64.
analyzer = DataAnalyzer('ThetaEstimate_4x100_N=1000')
sample_id = 32
regime_label = base_name + '%d'%theta
file_name = base_name + '_%d'%sample_id
print file_name
binnedTrain = BinnedSpikeTrain.initFromFile(file_name, phi_norms)
binnedTrain.pruneBins(None, N_thresh = 1, T_thresh=T_thresh)
regime_name = 'theta%d'%theta
abg_true = analyzer.getTrueParamValues(regime_name);
print abg_true
abg_fortet = analyzer.getEstimates(sample_id, regime_name, 'Fortet')[0]
print abg_fortet
visualizeData_vs_Fortet(abg_fortet, binnedTrain, theta,title_tag = 'Fortet: estimates', save_fig_name='theta20_Fortet_estimates')
visualizeData_vs_Fortet(abg_true, binnedTrain, theta,title_tag = 'Fortet: true', save_fig_name='theta20_Fortet_true')
abg_fp = analyzer.getEstimates(sample_id, regime_name, 'FP')[0]
print abg_fp
dx = .025; dt = FPMultiPhiSolver.calculate_dt(dx, abg_true, -1.0)
phis = binnedTrain.bins.keys();
S = FPMultiPhiSolver(binnedTrain.theta, phis,
dx, dt,
binnedTrain.getTf(), X_min = -1.0)
visualizeData_vs_FP(S, abg_fp, binnedTrain,title_tag = 'FP: estimates', save_fig_name='theta20_FP_estimates')
visualizeData_vs_FP(S, abg_true, binnedTrain,title_tag = 'FP: true', save_fig_name='theta20_FP_true')
def Harness(sample_id=13, regime_name='superSin', N_spikes = 1000, visualize=False):
from scipy.stats.distributions import norm
N_phi = 20;
phi_norms = linspace(1/(2.*N_phi), 1. - 1/ (2.*N_phi), N_phi)
base_name = 'sinusoidal_spike_train_N=%d_'%N_spikes
regime_label = base_name + regime_name
# T_thresh = 128.;
file_name = regime_label + '_' + str(sample_id)
print file_name
binnedTrain = BinnedSpikeTrain.initFromFile(file_name, phi_norms)
# print 'Warning: pruning bins'
# binnedTrain.pruneBins(None, N_thresh = 100)
bins = binnedTrain.bins;
phis = bins.keys()
N_phi = len(phis)
alpha, beta, gamma, theta = binnedTrain._Train._params.getParams()
def loss_function_simple(abg, visualize, fig_tag = ''):
error = .0;
if visualize:
figure()
for (phi_m, phi_idx) in zip(phis, xrange(N_phi)):
Is = bins[phi_m]['Is']
uniqueIs = bins[phi_m]['unique_Is']
a,b,g = abg[0], abg[1], abg[2]
movingThreshold = getMovingThreshold(a,g, phi_m,binnedTrain.theta)
LHS_numerator = movingThreshold(uniqueIs[1:]) *sqrt(2.)
LHS_denominator = b * sqrt(1 - exp(-2*uniqueIs[1:]))
LHS = 1 - norm.cdf(LHS_numerator / LHS_denominator)
RHS = zeros_like(LHS)
N = len(Is)
for rhs_idx in xrange(1,len(uniqueIs)):
t = uniqueIs[rhs_idx]
lIs = Is[Is<t]
taus = t - lIs;
numerator = (movingThreshold(t) - movingThreshold(lIs)* exp(-taus)) * sqrt(2.)
denominator = b * sqrt(1. - exp(-2*taus))
RHS[rhs_idx-1] = sum(1. - norm.cdf(numerator/denominator)) / N
weight = len(Is)
lerror = dot((LHS - RHS)**2 , diff(uniqueIs)) * weight;
error += lerror
if visualize:
subplot(ceil(len(phis)/2),2, phi_idx+1);hold(True)
ts = uniqueIs[1:];
plot(ts, LHS, 'b');
plot(ts, RHS, 'rx');
# annotate('$\phi$ = %.2g'%(phi_m), ((min(ts), max(LHS)/2.)), )
annotate('lerror = %.3g'%lerror,((min(ts), max(LHS)/2.)), )
if visualize:
subplot(ceil(len(phis)/2),2, 1);
title(fig_tag)
return error
def loss_function_nonvectorized(abg, visualize=False):
error = .0;
if visualize:
figure();
for (phi_m, phi_idx) in zip(phis, xrange(N_phi)):
# for phi_m in phis:
Is = bins[phi_m]['Is']
N = len(Is);
Tf = amax(Is);
a,b,g = abg[0], abg[1], abg[2]
movingThreshold = getMovingThreshold(a,g, phi_m)
def LHS(ts):
LHS_numerator = movingThreshold(ts) *sqrt(2.)
LHS_denominator = b * sqrt(1 - exp(-2*ts))
return 1 - norm.cdf(LHS_numerator / LHS_denominator);
# def RHS(t):
# lIs = Is[Is<t];
# taus = t - lIs;
# numerator = (movingThreshold(t) - movingThreshold(lIs)* exp(-taus)) * sqrt(2.)
# denominator = b * sqrt(1. - exp(-2*taus))
# return sum(1. - norm.cdf(numerator/denominator)) / N
def RHS(ts):
if False == iterable(ts):
ts = [ts]
rhs = empty_like(ts)
for t, t_indx in zip(ts, xrange(size(ts))):
lIs = Is[Is<t];
taus = t - lIs;
numerator = (movingThreshold(t) - movingThreshold(lIs)* exp(-taus)) * sqrt(2.)
denominator = b * sqrt(1. - exp(-2*taus))
rhs[t_indx] = sum(1. - norm.cdf(numerator/denominator)) / N
return rhs
integrand = lambda t: (LHS(t) - RHS(t)) **2
from scipy.integrate import quad, quadrature, fixed_quad
# quadrature, quad_error = quad(integrand, a= 1e-8, b=Tf+1e-8, limit = 50)
quadrature, quad_error = quadrature(integrand, a= 1e-8, b=Tf+1e-8,
tol=5e-03, rtol=1.49e-04,
maxiter = 64,
vec_func = True)
# val , err_msg = fixed_quad( integrand, a= 1e-8, b=Tf+1e-8,
# n = 12)
# print 'quadrature = ',quadrature
# print 'val = ',val
# print 'difference = ', quadrature - val
weight = len(Is)
#VISUALIZE FOR NOW:
if visualize:
subplot(ceil(len(phis)/2),2, phi_idx+1);hold(True)
ts = linspace(1e-8, Tf+1e-8, 100) ;
lhs = empty_like(ts); rhs = empty_like(ts);
for t, t_indx in zip(ts,
xrange(len(ts))):
lhs[t_indx] = LHS(t);
rhs[t_indx] = RHS(t);
plot(ts, lhs, 'b');
plot(ts, rhs, 'rx');
error += quadrature* weight;
return error
def loss_function_quadGaussian(abg, visualize=False, fig_tag = ''):
error = .0;
if visualize:
figure();
for (phi_m, phi_idx) in zip(phis, xrange(N_phi)):
# for phi_m in phis:
Is = bins[phi_m]['Is']
unique_Is = bins[phi_m]['unique_Is']
N_Is = len(Is);
Tf = amax(Is);
a,b,g = abg[0], abg[1], abg[2]
movingThreshold = getMovingThreshold(a,g, phi_m)
def LHS(ts):
LHS_numerator = movingThreshold(ts) *sqrt(2.)
LHS_denominator = b * sqrt(1 - exp(-2*ts))
return 1 - norm.cdf(LHS_numerator / LHS_denominator);
def RHS(ts):
if False == iterable(ts):
ts = array([ts])
# rhs = empty_like(ts)
# Is.reshape((len(Is),1))
lIs = tile(Is, len(ts) ).reshape((len(ts), len(Is))).transpose()
lts = tile(ts, (len(Is),1 ) )
mask = lIs < lts
taus = (lts - lIs); #*mask
#NOTE BELOW WE use abs(taus) since for non-positive taus we will mask away anyway:
numerator = (movingThreshold(lts) - movingThreshold(lIs)* exp(-abs(taus))) * sqrt(2.)
denominator = b * sqrt(1. - exp(-2*abs(taus)))
rhs = sum( (1. - norm.cdf(numerator/denominator))*mask, axis=0) / N_Is
return rhs
integrand = lambda t: (LHS(t) - RHS(t)) **2
from scipy.integrate import quad, quadrature, fixed_quad
# valcheck, quad_error = quad(integrand, a= 1e-8, b=Tf+1e-8, limit = 64)
val, quad_error = quadrature(integrand, a= 1e-8, b=Tf+1e-8,
tol=5e-03, rtol=1.49e-04,
maxiter = 64,
vec_func = True)
weight = len(Is)
lerror = val* weight;
error += lerror
if visualize:
subplot(ceil(len(phis)/2),2, phi_idx+1); hold(True)
# ts = linspace(1e-8, Tf+1e-8, 100) ;
ts = unique_Is[1:];
lhs = LHS(ts);
rhs = RHS(ts);
plot(ts, lhs, 'b');
plot(ts, rhs, 'rx');
annotate('lerror = %.3g'%lerror,((min(ts), max(lhs)/2.)), )
if visualize:
subplot(ceil(len(phis)/2),2,1);
title(fig_tag)
return error
def loss_function_L1(abg, visualize=False, fig_tag = ''):
error = .0;
if visualize:
figure();
for (phi_m, phi_idx) in zip(phis, xrange(N_phi)):
# for phi_m in phis:
Is = bins[phi_m]['Is']
unique_Is = bins[phi_m]['unique_Is']
N_Is = len(Is);
Tf = amax(Is);
a,b,g = abg[0], abg[1], abg[2]
movingThreshold = getMovingThreshold(a,g, phi_m, theta)
def LHS(ts):
LHS_numerator = movingThreshold(ts) *sqrt(2.)
LHS_denominator = b * sqrt(1 - exp(-2*ts))
return 1 - norm.cdf(LHS_numerator / LHS_denominator);
def RHS(ts):
if False == iterable(ts):
ts = array([ts])
# rhs = empty_like(ts)
# Is.reshape((len(Is),1))
lIs = tile(Is, len(ts) ).reshape((len(ts), len(Is))).transpose()
lts = tile(ts, (len(Is),1 ) )
mask = lIs < lts
taus = (lts - lIs); #*mask
#NOTE BELOW WE use abs(taus) since for non-positive taus we will mask away anyway:
numerator = (movingThreshold(lts) - movingThreshold(lIs)* exp(-abs(taus))) * sqrt(2.)
denominator = b * sqrt(1. - exp(-2*abs(taus)))
rhs = sum( (1. - norm.cdf(numerator/denominator))*mask, axis=0) / N_Is
return rhs
integrand = lambda t: abs(LHS(t) - RHS(t))
from scipy.integrate import quad, quadrature, fixed_quad
print unique_Is
val, quad_error = quad(integrand, a= 1e-8, b=Tf+1., limit = 64,
points = sort(unique_Is) )
# val, quad_error = quadrature(integrand, a= 1e-8, b=Tf+1e-8,
# tol=5e-03, rtol=1.49e-04,
# maxiter = 64,
# vec_func = True)
# val , err_msg = fixed_quad( integrand, a= 1e-8, b=Tf+1e-8,
weight = len(Is)
lerror = val* weight;
error += lerror
if visualize:
subplot(ceil(len(phis)/2),2, phi_idx+1); hold(True)
# ts = linspace(1e-8, Tf+1e-8, 100) ;
ts = unique_Is[1:];
lhs = LHS(ts);
rhs = RHS(ts);
plot(ts, lhs, 'b');
plot(ts, rhs, 'rx');
annotate('lerror = %.3g'%lerror,((min(ts), max(lhs)/2.)), )
if visualize:
subplot(ceil(len(phis)/2),2,1); title(fig_tag)
return error
def loss_function_manualquad(abg, visualize=False, fig_tag = ''):
error = .0;
if visualize:
figure();
for (phi_m, phi_idx) in zip(phis, xrange(N_phi)):
Is = bins[phi_m]['Is']
unique_Is = bins[phi_m]['unique_Is']
N_Is = len(Is);
Tf = amax(Is);
a,b,g = abg[0], abg[1], abg[2]
movingThreshold = getMovingThreshold(a,g, phi_m, binnedTrain.theta)
def LHS(ts):
LHS_numerator = movingThreshold(ts) *sqrt(2.)
LHS_denominator = b * sqrt(1 - exp(-2*ts))
return 1 - norm.cdf(LHS_numerator / LHS_denominator);
def RHS(ts):
if False == iterable(ts):
ts = array([ts])
lIs = tile(Is, len(ts) ).reshape((len(ts), len(Is))).transpose()
lts = tile(ts, (len(Is),1 ) )
mask = lIs < lts
taus = (lts - lIs); #*mask
#NOTE BELOW WE use abs(taus) since for non-positive taus we will mask away anyway:
numerator = (movingThreshold(lts) - movingThreshold(lIs)* exp(-abs(taus))) * sqrt(2.)
denominator = b * sqrt(1. - exp(-2*abs(taus)))
rhs = sum( (1. - norm.cdf(numerator/denominator))*mask, axis=0) / N_Is
return rhs
integrand = lambda t: (LHS(t) - RHS(t)) **2
dt = 1e-2;
ts_manual = arange(1e-8, Tf+1e-2, dt )
integrand = LHS(ts_manual) - RHS(ts_manual)
val = dot(integrand, integrand)*dt;
weight = len(Is)
lerror = val* weight;
error += lerror
if visualize:
subplot(ceil(len(phis)/2),2, phi_idx+1); hold(True)
# ts = linspace(1e-8, Tf+1e-8, 100) ;
ts = unique_Is[1:];
lhs = LHS(ts);
rhs = RHS(ts);
plot(ts, lhs, 'b');
plot(ts, rhs, 'rx');
annotate('lerror = %.3g'%lerror,((min(ts), max(lhs)/2.)), )
if visualize:
subplot(ceil(len(phis)/2),2,1);
title(fig_tag)
return error
def loss_function_supnormalized(abg, visualize=False, fig_tag = ''):
error = .0;
if visualize:
figure();
for (phi_m, phi_idx) in zip(phis, xrange(N_phi)):
Is = bins[phi_m]['Is']
unique_Is = bins[phi_m]['unique_Is']
N_Is = len(Is);
Tf = amax(Is);
a,b,g = abg[0], abg[1], abg[2]
movingThreshold = getMovingThreshold(a,g, phi_m, binnedTrain.theta)
def LHS(ts):
LHS_numerator = movingThreshold(ts) *sqrt(2.)
LHS_denominator = b * sqrt(1 - exp(-2*ts))
return 1 - norm.cdf(LHS_numerator / LHS_denominator);
def RHS(ts):
if False == iterable(ts):
ts = array([ts])
lIs = tile(Is, len(ts) ).reshape((len(ts), len(Is))).transpose()
lts = tile(ts, (len(Is),1 ) )
mask = lIs < lts
taus = (lts - lIs); #*mask
#NOTE BELOW WE use abs(taus) since for non-positive taus we will mask away anyway:
numerator = (movingThreshold(lts) - movingThreshold(lIs)* exp(-abs(taus))) * sqrt(2.)
denominator = b * sqrt(1. - exp(-2*abs(taus)))
rhs = sum( (1. - norm.cdf(numerator/denominator))*mask, axis=0) / N_Is
return rhs
dt = 1e-3;
ts_manual = arange(1e-8, Tf+1e-2, dt)
lhs = LHS(ts_manual)
difference = abs(lhs - RHS(ts_manual))/amax(lhs)
val = amax(difference);
weight = len(Is)
lerror = val* weight;
error += lerror
if visualize:
subplot(ceil(len(phis)/2),2, phi_idx+1); hold(True)
# ts = linspace(1e-8, Tf+1e-8, 100) ;
ts = unique_Is[1:];
lhs = LHS(ts);
rhs = RHS(ts);
plot(ts, lhs, 'b');
plot(ts, rhs, 'rx');
annotate('lerror = %.3g'%lerror,((min(ts), max(lhs)/2.)), )
if visualize:
subplot(ceil(len(phis)/2),2,1);
title(fig_tag)
return error
def loss_function_sup(abg, visualize=False, fig_tag = ''):
error = .0;
if visualize:
figure();
for (phi_m, phi_idx) in zip(phis, xrange(N_phi)):
Is = bins[phi_m]['Is']
unique_Is = bins[phi_m]['unique_Is']
N_Is = len(Is);
Tf = amax(Is);
a,b,g = abg[0], abg[1], abg[2]
movingThreshold = getMovingThreshold(a,g, phi_m, binnedTrain.theta)
def LHS(ts):
LHS_numerator = movingThreshold(ts) *sqrt(2.)
LHS_denominator = b * sqrt(1 - exp(-2*ts))
return 1 - norm.cdf(LHS_numerator / LHS_denominator);
def RHS(ts):
if False == iterable(ts):
ts = array([ts])
lIs = tile(Is, len(ts) ).reshape((len(ts), len(Is))).transpose()
lts = tile(ts, (len(Is),1 ) )
mask = lIs < lts
taus = (lts - lIs); #*mask
#NOTE BELOW WE use abs(taus) since for non-positive taus we will mask away anyway:
numerator = (movingThreshold(lts) - movingThreshold(lIs)* exp(-abs(taus))) * sqrt(2.)
denominator = b * sqrt(1. - exp(-2*abs(taus)))
rhs = sum( (1. - norm.cdf(numerator/denominator))*mask, axis=0) / N_Is
return rhs
dt = 1e-3;
ts_manual = arange(1e-8, Tf+1e-2, dt)
lhs = LHS(ts_manual)
difference = abs(lhs - RHS(ts_manual))
val = amax(difference);
weight = len(Is)
lerror = val* weight;
error += lerror
if visualize:
subplot(ceil(len(phis)/2),2, phi_idx+1); hold(True)
# ts = linspace(1e-8, Tf+1e-8, 100) ;
ts = unique_Is[1:];
lhs = LHS(ts);
rhs = RHS(ts);
plot(ts, lhs, 'b');
plot(ts, rhs, 'rx');
annotate('lerror = %.3g'%lerror,((min(ts), max(lhs)/2.)), )
if visualize:
subplot(ceil(len(phis)/2),2,1);
title(fig_tag)
return error
#EXPERIMENT:
# Analyzer = DataAnalyzer()
def outlinept():
pass
analyzer = DataAnalyzer('FvsWF_4x16');
abg_true = analyzer.getTrueParamValues(regime_name)
loss_function_L1(abg_true, visualize=True)
return
quad_estimated = analyzer.getEstimates(sample_id, regime_name, 'QuadFortet')[0]
simple_estimated = analyzer.getEstimates(sample_id, regime_name, 'Fortet')[0]
for abg, tag, L in zip(3*[abg_true, quad_estimated],
['sup_true_params' , 'sup_estimated_params',
'supnormailzed_true_params', 'supnormailzed_estimated_params',
'manualquad_true_params' , 'manualquad_estimated_params'],
2*[loss_function_sup]+
2*[loss_function_supnormalized] +
2*[loss_function_manualquad]):
start = time.clock()
loss = L(abg,visualize, fig_tag = regime + '_' + tag);
end = time.clock()
print tag, ':%.2f,%.2f,%.2f:' %(abg[0],abg[1],abg[2]), 'error = %.4f'%loss , ' | compute time = ', end - start
