def MC_comp_pfmethods_ML(self, shape):
mask = np.ones(shape, dtype=int) #full mask
g = graph_from_lattice(mask, kerMask=kerMask2D_4n, toroidal=True)
# Betas to iterate over ...
betaMax = 1.4
dbeta = .1
betas = np.arange(0., betaMax, dbeta)
# nb of monte carlo iterations
mcit = 40#100
# nb of classes
nbc = 2
# Path sampling:
mBePS, stdBePS, stdBeMCPS = cached_eval(test_beta_estim_obs_fields,
([g], betas, nbc, 'PS',
mcit), path=config.cacheDir,
prefix='pyhrf_',
new=False)
# Extrapolation scheme:
mBeES, stdBeES, stdBeMCES = cached_eval(test_beta_estim_obs_fields,
([g], betas, nbc, 'ES',
mcit), path=config.cacheDir,
prefix='pyhrf_',
new=False)
# Onsager:
#mBeON, stdBeON, stdBeMCON = cached_eval(test_beta_estim_obs_fields,
# ([g], betas, nbc, 'ONSAGER',
# mcit), path=config.cacheDir,
# prefix='pyhrf_',
# new=True)
if self.plot:
import matplotlib.pyplot as plt
plt.figure()
plt.errorbar(betas, mBeES, stdBeMCES, fmt='r', label='ES')
plt.errorbar(betas, mBePS, stdBeMCPS, fmt='b', label='PS')
#plt.errorbar(betas, mBeON, stdBeMCON, fmt='g', label='Onsager')
plt.plot(betas, betas, 'k', label='true')
#plt.xlabel('simulated beta')
#plt.ylabel('beta MAP')
plt.legend(loc='upper left')
plt.ylim((0.0,1.45))
#plt.title('MC validation (%d it) for beta estimation on '\
# ' observed fields.' %mcit)
#fn = 'comp_obs_field_MC_PS_ES_Onsager_%dx%d' %shape
fn = 'comp_obs_field_MC_PS_%dx%d' %shape
figFn = os.path.join(self.outDir, figfn(fn))
plt.savefig(figFn)
def test_beta_estim_jde_mcmc(self):
nbCond = 2
betaMax = 1.2
dBeta = .1
betas = np.arange(0, betaMax, dBeta)
height = 20
width = 20
graph = graph_from_lattice(np.ones((height,width),dtype=int),
kerMask2D_4n)
gridLnZ = cached_eval(Cpt_Vec_Estim_lnZ_Graph, (graph, 2))
nbIt = 3
bolds = []
outputs = []
for beta in betas:
labels = [genPotts(graph, beta).reshape((height,width))
for i in range(nbCond)]
b = simulate_bold_from_labels(labels)
bolds.append(b)
p = {'data':b.to_fmri_data(np.ones((height,width),dtype=int)),
'nbIterations':nbIt,
'betaSimu':beta,
#'noiseVarSimu':noiseVar,
'gridLnZ':gridLnZ}
#outs = cached_eval(runEstimationBetaEstim, (p,))
#outputs.append(outs)
#view_results(bolds, outputs, betas)
scm = "0;0;0.333333;1;0.651282;0;1;1#" \
"0;0;0.333333;1;0.661538;0;1;0#" \
"0;0;0.333333;1;0.664103;1;1;0"
cm = cmstring_to_mpl_cmap(scm)
import plt
for output, beta in zip(outputs,betas):
lmap = output['pm_Labels_marked'].data[0,:,:,0]
plt.figure()
img = plt.matshow(lmap, cmap=cm)
img.get_axes().set_axis_off()
sbeta = ('%1.2f'%beta).replace('.','_')
plt.savefig('./hidden_field_pm_labels_marked_'+sbeta)
def test_slow_func(self):
t0 = time.time()
cached_eval(slow_func, {'a': 4, 'b': 6}, path=self.cache_dir)
delta = time.time() - t0
t0 = time.time()
cached_eval(slow_func, {'a': 4, 'b': 6}, path=self.cache_dir)
delta = time.time() - t0
assert delta < .1
t0 = time.time()
cached_eval(slow_func, {'a': 4, 'b': 8}, path=self.cache_dir)
delta = time.time() - t0
def test_slow_func(self):
t0 = time.time()
cached_eval(slow_func, {'a': 4, 'b': 6}, path=self.cache_dir)
delta = time.time() - t0
t0 = time.time()
cached_eval(slow_func, {'a': 4, 'b': 6}, path=self.cache_dir)
delta = time.time() - t0
assert delta < .1
t0 = time.time()
cached_eval(slow_func, {'a': 4, 'b': 8}, path=self.cache_dir)
delta = time.time() - t0
def beta_estim_obs_field_mc(graph, nbClasses, beta, gridLnz, mcit=1,
cachePotts=False):
cumulBetaEstim = 0.
cumul2BetaEstim = 0.
for mci in xrange(mcit):
labels = cached_eval(genPotts, (graph, beta, nbClasses),
save=cachePotts, new=(not cachePotts))
print 'labels:',labels
betaEstim, _ = beta_estim_obs_field(graph, labels, gridLnz,
method='ML')
#betaEstim, pb = beta_estim_obs_field(graph, labels, gridLnz,
# method='MAP')
print ' -mc- be=%f' %betaEstim
cumulBetaEstim += betaEstim
cumul2BetaEstim += betaEstim**2
meanBetaEstim = cumulBetaEstim / mcit
stdBetaEstim = ((cumul2BetaEstim / mcit) - meanBetaEstim**2)**.5
return meanBetaEstim, stdBetaEstim
def test_code_digest(self):
t0 = time.time()
cached_eval(slow_func, {'a': 4, 'b': 8}, digest_code=True,
path=self.cache_dir)
delta = time.time() - t0
def test_simple_args(self):
cached_eval(foo_func, {'a': 4, 'b': 6}, path=self.cache_dir)
def test_simple(self):
cached_eval(foo_func, (1, 2), path=self.cache_dir)
def test_beta_estim_obs_fields(graphs, betas, nbLabels, pfmethod, mcit=1):
nbGraphs = len(graphs)
meanBetaEstims = np.zeros(len(betas))
stdBetaEstims = np.zeros(len(betas))
stdMCBetaEstims = np.zeros(len(betas))
gridLnz = range(nbGraphs)
if mcit == 1:
cachePotts = True
else:
cachePotts = False
# compute partition functions:
print 'compute partition functions ... nbgraphs:', nbGraphs
print 'Method =', pfmethod
print 'nbLabels =', nbLabels
if pfmethod == 'ES':
for ig, g in enumerate(graphs):
grid = cached_eval(Cpt_Vec_Estim_lnZ_Graph_fast3, (g,nbLabels),
path=config.cacheDir, prefix='pyhrf_')
gridLnz[ig] = (grid[0][:-1], grid[1][:-1])
elif pfmethod == 'PS':
for ig, g in enumerate(graphs):
gridLnz[ig] = cached_eval(Cpt_Vec_Estim_lnZ_Graph, (g,nbLabels),
path=config.cacheDir, new=False,
prefix='pyhrf_')
elif pfmethod == 'ONSAGER':
for ig, g in enumerate(graphs):
betaGrid = cached_eval(Cpt_Vec_Estim_lnZ_Graph,([[]],nbLabels),
path=config.cacheDir)[1]
lpf = cached_eval(logpf_ising_onsager, (len(g),betaGrid),
path=config.cacheDir, prefix='pyhrf_')
gridLnz[ig] = (lpf, betaGrid)
else:
print 'unknown pf method!'
sys.exit(1)
print 'gridLnz:'
print gridLnz
# Beta estimation:
print 'beta estimation test loop ...'
for ib, b in enumerate(betas):
cumulBetaEstim = 0.
cumul2BetaEstim = 0.
cumulStdMC = 0.
print 'testing with beta =', b, '...'
for ig, graph in enumerate(graphs):
mbe, stdmc = beta_estim_obs_field_mc(graph, nbLabels, b,
gridLnz[ig], mcit,
cachePotts=False)
print '%f(%f)' %(mbe,stdmc)
cumulBetaEstim += mbe
cumulStdMC += stdmc
cumul2BetaEstim += mbe**2
#print betaEstim, cumulBetaEstim, cumul2BetaEstim
meanBetaEstims[ib] = cumulBetaEstim / nbGraphs
stdBetaEstims[ib] = ((cumul2BetaEstim / nbGraphs) - \
meanBetaEstims[ib]**2)**.5
stdMCBetaEstims[ib] = cumulStdMC / nbGraphs
#print '-> ', meanBetaEstims[ib], stdBetaEstims[ib]
return meanBetaEstims, stdBetaEstims, stdMCBetaEstims
def test_code_digest(self):
t0 = time.time()
cached_eval(slow_func, {'a': 4, 'b': 8}, digest_code=True,
path=self.cache_dir)
delta = time.time() - t0
def test_simple_args(self):
cached_eval(foo_func, {'a': 4, 'b': 6}, path=self.cache_dir)
def test_simple(self):
cached_eval(foo_func, (1, 2), path=self.cache_dir)
def MC_comp_pfmethods_ML_3C(self, shape):
newEval = config.updateCache
mask = np.ones(shape, dtype=int) #full mask
g = graph_from_lattice(mask, kerMask=kerMask2D_4n, toroidal=True)
# Betas to iterate over ...
betaMax = 1.4
dbeta = .1
betas = np.arange(0., betaMax, dbeta)
# nb of monte carlo iterations
mcit = 100#100
# nb of classes
nbc = 3
# Path sampling:
mBePS, stdBePS, stdBeMCPS = cached_eval(test_beta_estim_obs_fields,
([g], betas, nbc, 'PS',
mcit), path=config.cacheDir,
prefix=pyhrf.tmpPrefix,
new=False)
# Extrapolation scheme:
mBeES, stdBeES, stdBeMCES = cached_eval(test_beta_estim_obs_fields,
([g], betas, nbc, 'ES',
mcit), path=config.cacheDir,
prefix=pyhrf.tmpPrefix,
new=False)
# Onsager:
#mBeON, stdBeON, stdBeMCON = cached_eval(test_beta_estim_obs_fields,
# ([g], betas, nbc, 'ONSAGER',
# mcit), path=config.cacheDir,
# prefix=pyhrf.tmpPrefix,
# new=True)
if self.plot:
import plt
plt.figure()
plt.errorbar(betas, mBeES, stdBeMCES, fmt='r', label='ES')
plt.errorbar(betas, mBePS, stdBeMCPS, fmt='b', label='PS')
#plt.errorbar(betas, mBeON, stdBeMCON, fmt='g', label='Onsager')
plt.plot(betas, betas, 'k', label='true')
#plt.xlabel('simulated beta')
#plt.ylabel('beta MAP')
plt.legend(loc='upper left')
#plt.title('MC validation (%d it) for beta estimation on '\
# ' observed fields.' %mcit)
fn = 'comp_obs_field_3class_MC_PS_ES_%dx%d' %shape
figFn = os.path.join(self.outDir, figfn(fn))
plt.savefig(figFn)
plt.figure()
grid = cached_eval(Cpt_Vec_Estim_lnZ_Graph_fast3, (g,nbc),
path=config.cacheDir, prefix=pyhrf.tmpPrefix)
lnzES, betas = (grid[0][:-1], grid[1][:-1])
lnzPS, betas = cached_eval(Cpt_Vec_Estim_lnZ_Graph, (g,nbc),
path=config.cacheDir, new=False,
prefix=pyhrf.tmpPrefix)
plt.plot(betas, lnzES, 'r', label='ES')
plt.plot(betas, lnzPS, 'b', label='PS')
fn = 'comp_PF_3class_PS_ES_%dx%d' %shape
figFn = os.path.join(self.outDir, figfn(fn))
plt.savefig(figFn)
