def compute_bootstrap_sample_components(x):
gf, Urd = x
# commo operation - generate new bootstrap sample
b = gf.sample_temporal_bootstrap()
# custom method to compute the components
Ur = COMPONENT_ESTIMATOR(b)
# match, flip sign and permute the discovered components
perm, sign_flip = match_components_munkres(Urd, Ur)
Ur = Ur[:, perm]
Ur *= sign_flip[:Ur.shape[1]]
return Ur
def compute_bootstrap_sample_components(x):
gf, Urd = x
# commo operation - generate new bootstrap sample
b = gf.sample_temporal_bootstrap()
# custom method to compute the components
Ur = COMPONENT_ESTIMATOR(b)
# match, flip sign and permute the discovered components
perm, sign_flip = match_components_munkres(Urd, Ur)
Ur = Ur[:, perm]
Ur *= sign_flip[:Ur.shape[1]]
return Ur
def compute_bootstrap_sample_components(x):
gf, Urd = x
Nc = Urd.shape[1]
# common operation - generate new bootstrap sample
b = gf.sample_temporal_bootstrap()
if COSINE_REWEIGHTING:
b *= gf.qea_latitude_weights()
# custom method to compute the components
Ur = COMPONENT_ESTIMATOR(b)
if Ur is None:
return None
# match, flip sign and permute the discovered components
perm, sign_flip = match_components_munkres(Urd, Ur)
Ur = Ur[:, perm[:Nc]]
Ur *= sign_flip[:, :Nc]
return Ur
def compute_bootstrap_sample_components(x):
gf, Urd = x
Nc = Urd.shape[1]
# common operation - generate new bootstrap sample
b = gf.sample_temporal_bootstrap()
if COSINE_REWEIGHTING:
b *= gf.qea_latitude_weights()
# custom method to compute the components
Ur = COMPONENT_ESTIMATOR(b)
if Ur is None:
return None
# match, flip sign and permute the discovered components
perm, sign_flip = match_components_munkres(Urd, Ur)
Ur = Ur[:, perm[:Nc]]
Ur *= sign_flip[:, :Nc]
return Ur
def compute_bootstrap_sample_components(gf, Urd, jobq, resq):
"""
Estimate the components from a temporally bootstrapped dataset.
"""
Nc = Urd.shape[1]
while jobq.get() is not None:
# common operation - generate new bootstrap sample
b, ndx = gf.sample_temporal_bootstrap()
if COSINE_REWEIGHTING:
b *= gf.qea_latitude_weights()
# custom method to compute the components
Res = COMPONENT_ESTIMATOR(b)
if Res is None:
resq.put(None)
else:
Ur, _ = Res
# match, flip sign and permute the discovered components
perm, sign_flip = match_components_munkres(Urd, Ur)
Ur = Ur[:, perm[:Nc]]
Ur *= sign_flip[:, :Nc]
resq.put(Ur)
def synth_model_plot_component_matching():
with open('results/synth_model_component_robustness_190smp_4comps.bin', 'r') as f:
d = cPickle.load(f)
Uopt = d['Uopt']
Ur = d['Ur']
mean_comp = d['mean']
var_comp = d['var']
perm, sf = match_components_munkres(Uopt, Ur)
perm = perm[:3]
sf = sf[:, :3]
rst = np.setdiff1d(np.arange(Ur.shape[1]), np.sort(perm[:3]), False)
print rst
Urm = np.hstack([Ur[:, perm] * sf, Ur[:, rst]])
print Urm.shape
perm, sf = match_components_munkres(Uopt, mean_comp)
perm = perm[:3]
sf = sf[:, :3]
rst = np.setdiff1d(np.arange(mean_comp.shape[1]), np.sort(perm[:3]), False)
print rst
Umvm = np.hstack([mean_comp[:, perm] * sf, mean_comp[:, rst]])
Umvv = np.hstack([var_comp[:, perm], var_comp[:, rst]])
print Umvm.shape
print Umvv.shape
p2 = list(perm)
for i in range(mean_comp.shape[1]):
if not i in p2:
p2.append(i)
print estimate_snr(Uopt, Urm[:, :3])
print estimate_snr(Uopt, Umvm[:, :3])
print mse_error(Uopt, Urm[:, :3])
print mse_error(Uopt, Umvm[:, :3])
print marpe_error(Uopt, Urm[:, :3])
print marpe_error(Uopt, Umvm[:, :3])
print np.sum(Umvv, axis = 0)
print np.sum(Umvm ** 2, axis = 0)
print np.sum(Umvm ** 2, axis = 0) / np.sum(Umvv[:, p2], axis = 0)
plt.figure(figsize = (12, 16))
plt.subplots_adjust(left = 0.05, bottom = 0.05, right = 0.95, top = 0.95)
for i in range(4):
plt.subplot(4, 3, i*3+1)
if(i < 3):
plt.plot(Uopt[:,i], 'r-')
plt.plot(Umvm[:,i], 'b-')
plt.plot(np.ones((Uopt.shape[0],1)) * (1.0 / Uopt.shape[0]**0.5), 'g-', linewidth = 2)
plt.title('Component %d [bootstrap mean]' % (i+1))
plt.subplot(4, 3, i*3+2)
if(i < 3):
plt.plot(Uopt[:,i], 'r-')
plt.plot(Urm[:,i], 'b-')
plt.plot(np.ones((Uopt.shape[0],1)) * (1.0 / Uopt.shape[0]**0.5), 'g-', linewidth = 2)
plt.title('Component %d [data]' % (i+1))
plt.subplot(4, 3, i*3+3)
plt.plot(Umvv[:,i], 'b-')
plt.title('Component %d [variance]' % (i+1))
plt.show()
