def run_svm_inference(self, test, w):
logger.info('running inference on: {}'.format(test))
outdir = self.dir_inf + test
if not os.path.isdir(outdir):
os.makedirs(outdir)
## segment test image with trained w
def wwf(im, _w):
''' meta weight function'''
data = 0
for iwf, wf in enumerate(self.weight_functions.values()):
ij, _data = wf(im)
data += _w[iwf] * _data
return ij, data
## load images and ground truth
file_seg = self.dir_reg + test + 'seg.hdr'
file_im = self.dir_reg + test + 'gray.hdr'
im = io_analyze.load(file_im)
seg = io_analyze.load(file_seg)
seg.flat[~np.in1d(seg.ravel(), self.labelset)] = self.labelset[0]
## save image
im = im / np.std(im) # normalize image by variance
## prior
anchor_api = BaseAnchorAPI(
self.prior,
anchor_weight=w[-1],
)
sol, y = rwsegment.segment(im,
anchor_api,
seeds=self.seeds,
weight_function=lambda im: wwf(im, w),
**self.rwparams_inf)
np.save(outdir + 'y.test.npy', y)
io_analyze.save(outdir + 'sol.test.hdr', sol.astype(np.int32))
## compute Dice coefficient
dice = compute_dice_coef(sol, seg, labelset=self.labelset)
np.savetxt(outdir + 'dice.test.txt',
np.c_[dice.keys(), dice.values()],
fmt='%d %.8f')
## inference compare with gold standard
dice_gold = np.loadtxt(outdir + 'dice.gold.txt')
y_gold = np.load(outdir + 'y.gold.npy')
sol_gold = io_analyze.load(outdir + 'sol.gold.hdr')
np.testing.assert_allclose(dice.values(),
dict(dice_gold).values(),
err_msg='FAIL: dice coef mismatch',
atol=1e-8)
np.testing.assert_allclose(y, y_gold, err_msg='FAIL: y mismatch')
np.testing.assert_equal(sol, sol_gold, err_msg='FAIL: sol mismatch')
print 'PASS: inference tests'
def load_or_compute_classifier(self,train,test,mask=None):
from rwsegment import boundary_utils
reload(boundary_utils)
#idir_prior = config.dir_prior_edges + test
#if not os.path.isdir(dir_prior):
# os.makedirs(dir_prior)
## Train classifier
logger.info('train classifier with train {} for test {}'.format(train, test))
##load image and seg
im = io_analyze.load(
config.dir_reg + test + train + 'reggray.hdr').astype(float)
nim = im/np.std(im)
seg = io_analyze.load(config.dir_reg + test + train + 'regseg.hdr')
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## sample points
points = boundary_utils.sample_points(im, self.step, mask=mask)
logger.debug('number of sampled points = {}'.format(len(points)))
#impoints = np.zeros(im.shape,dtype=int)
#impoints[tuple(points.T)] = np.arange(len(points)) + 1
## compute edges
edges,edgev,labels = boundary_utils.get_edges(im, points, mask=mask)
logger.debug('number of edges = {}'.format(len(edges)))
## extract profiles
profiles,emap,dists = boundary_utils.get_profiles(nim, points, edges, rad=0)
logger.debug('extracted profiles')
## make features
x = boundary_utils.make_features(profiles, size=self.sizex, additional=[dists,edgev,edgev/dists])
logger.debug('features made, size = {}'.format(len(x[0])))
## make annotations
z = boundary_utils.is_boundary(points, edges, seg)
logger.debug('annotations made')
## learn profiles
logger.debug('training classifier')
classifier = boundary_utils.Classifier()
classifier.train(x,z)
## test classification
logger.debug('testing classifier')
cl, scores = classifier.classify(x)
logger.info('non boundary correct rate: {:.3}'.format(
np.sum((np.r_[cl]==0)&(np.r_[z]==0))/np.sum(np.r_[z]==0).astype(float)))
logger.info('boundary correct rate: {:.3}'.format(
np.sum((np.r_[cl]==1)&(np.r_[z]==1))/np.sum(np.r_[z]==1).astype(float)))
## store classifier
#np.savetxt(dir_prior + 'classifier.txt', classifier.w)
return classifier
def process_sample(self, test, fold=None):
## get prior
prior, mask = load_or_compute_prior_and_mask(
test, force_recompute=self.force_recompute_prior, pca=True, fold=fold
)
seeds = (-1) * mask
mask = mask.astype(bool)
## load image
file_name = config.dir_reg + test + "gray.hdr"
logger.info("segmenting data: {}".format(file_name))
im = io_analyze.load(file_name)
file_gt = config.dir_reg + test + "seg.hdr"
seg = io_analyze.load(file_gt)
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## normalize image
nim = im / np.std(im)
## init anchor_api
anchor_api = MetaAnchor(
prior=prior, prior_models=self.prior_models, prior_weights=self.prior_weights, image=nim
)
## start segmenting
# import ipdb; ipdb.set_trace()
sol, impca = rwsegment_pca.segment(nim, anchor_api, seeds=seeds, labelset=self.labelset, **self.params)
## compute Dice coefficient per label
dice = compute_dice_coef(sol, seg, labelset=self.labelset)
logger.info("Dice: {}".format(dice))
dice_pca = compute_dice_coef(impca, seg, labelset=self.labelset)
logger.info("Dice pca only: {}".format(dice_pca))
if not config.debug:
if fold is not None:
test_name = "f{}_{}".format(fold[0][:2], test)
else:
test_name = test
outdir = config.dir_seg + "/{}/{}".format(self.model_name, test_name)
logger.info("saving data in: {}".format(outdir))
if not os.path.isdir(outdir):
os.makedirs(outdir)
io_analyze.save(outdir + "sol.hdr", sol.astype(np.int32))
io_analyze.save(outdir + "solpca.hdr", impca.astype(np.int32))
np.savetxt(outdir + "dice.txt", np.c_[dice.keys(), dice.values()], fmt="%d %.8f")
np.savetxt(outdir + "dice_pca.txt", np.c_[dice.keys(), dice_pca.values()], fmt="%d %.8f")
def batch_compute_dice(folder_list, labelset=[13,14,15,16]):
import config
dir_gt = config.dir_reg
lbset = np.asarray(labelset, dtype=int)
for folder in folder_list:
for vol in config.vols:
if not os.path.isfile(folder + '/' + vol + '/sol.hdr'):
continue
print 'computing dice for segmentation: {}'.format(folder + '/' + vol + '/sol.hdr')
sol = io_analyze.load(folder + '/' + vol + '/sol.hdr').astype(int)
gt = io_analyze.load(dir_gt + '/' + vol + '/seg.hdr').astype(int)
d_slice = compute_dice_per_slice(sol, gt, labelset=labelset)
d_labels = compute_dice_coef(sol, gt, labelset=labelset)
#print d_slice, d_labels
np.savetxt(folder + '/' + vol + '/dice_labels.txt', np.c_[d_labels.keys(), d_labels.values()], fmt='%d %f')
np.savetxt(folder + '/' + vol + '/dice_slices.txt', np.c_[d_slice.keys(), d_slice.values()], fmt='%d %f')
def compute_mask(self, train):
seg = io_analyze.load(config.dir_reg + test + train + 'regseg.hdr')
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
import ndimage
struct = np.ones((20, ) * mask.ndim)
mask = ndimage.binary_dilation(
seg > 0,
structure=struct,
)
return mask
def compute_mask(self, train):
seg = io_analyze.load(config.dir_reg + test + train + 'regseg.hdr')
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
import ndimage
struct = np.ones((20,)*mask.ndim)
mask = ndimage.binary_dilation(
seg>0,
structure=struct,
)
return mask
def train_svm(self,test):
outdir = test
## training images and segmentations
self.training_set = []
for train in self.training_vols:
if test==train: continue
logger.info('loading training data: {}'.format(train))
file_seg = self.dir_reg + test + train + 'regseg.hdr'
file_im = self.dir_reg + test + train + 'reggray.hdr'
im = io_analyze.load(file_im)
im = im/np.std(im) # normalize image by std
seg = io_analyze.load(file_seg)
seg.flat[~np.in1d(seg.ravel(),self.labelset)] = self.labelset[0]
z = (seg.ravel()==np.c_[self.labelset])# make bin vector z
self.training_set.append((im, z))
## instanciate functors
self.svm_rwmean_api = SVMRWMeanAPI(
self.prior,
self.weight_functions,
self.labelset,
self.rwparams_svm,
seeds=self.seeds,
)
## learn struct svm
logger.debug('start learning')
self.svm = struct_svm.StructSVM(
self.training_set,
self.svm_rwmean_api.compute_loss,
self.svm_rwmean_api.compute_psi,
self.svm_rwmean_api.compute_mvc,
**self.svmparams
)
w,xi,info = self.svm.train()
return w,xi,info
def train_svm(self, test):
outdir = test
## training images and segmentations
self.training_set = []
for train in self.training_vols:
if test == train: continue
logger.info('loading training data: {}'.format(train))
file_seg = self.dir_reg + test + train + 'regseg.hdr'
file_im = self.dir_reg + test + train + 'reggray.hdr'
im = io_analyze.load(file_im)
im = im / np.std(im) # normalize image by std
seg = io_analyze.load(file_seg)
seg.flat[~np.in1d(seg.ravel(), self.labelset)] = self.labelset[0]
z = (seg.ravel() == np.c_[self.labelset]) # make bin vector z
self.training_set.append((im, z))
## instanciate functors
self.svm_rwmean_api = SVMRWMeanAPI(
self.prior,
self.weight_functions,
self.labelset,
self.rwparams_svm,
seeds=self.seeds,
)
## learn struct svm
logger.debug('start learning')
self.svm = struct_svm.StructSVM(self.training_set,
self.svm_rwmean_api.compute_loss,
self.svm_rwmean_api.compute_psi,
self.svm_rwmean_api.compute_mvc,
**self.svmparams)
w, xi, info = self.svm.train()
return w, xi, info
def load_or_compute_orientations(self, train, test, mask=None):
from rwsegment import boundary_utils
reload(boundary_utils)
## Train classifier
logger.info('train orientations with train {} for test {}'.format(
train, test))
##load image and seg
seg = io_analyze.load(config.dir_reg + test + train +
'regseg.hdr').astype(int)
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## sample points
points = boundary_utils.sample_points(np.ones(seg.shape), 5, mask=mask)
logger.debug('number of sampled points = {}'.format(len(points)))
## find edges between muscles
points_label = seg[tuple(points.T)]
nlabel = len(self.labelset)
hist = {}
orient_scores = np.zeros((self.orients.shape[0], self.nlabel**2))
#orient_prior = np.ones((3,self.label_pairs.shape[0]))
ipair = 0
for l1 in range(nlabel):
label1 = self.labelset[l1]
inds1 = np.where(points_label == label1)[0]
hist[label1] = {}
for l2 in range(nlabel):
label2 = self.labelset[l2]
inds2 = np.where(points_label == label2)[0]
edges = np.argwhere(
np.triu(np.ones((inds1.size, inds2.size)), k=1))
edges = np.c_[inds1[edges[:, 0]], inds2[edges[:, 1]]]
vecs = points[edges[:, 1]] - points[edges[:, 0]]
vecs = vecs / np.c_[np.sqrt(np.sum(vecs**2, axis=1))]
if l1 == 0 or l1 == l2:
avg = np.ones(len(self.orients)) / float(len(self.orients))
else:
scores = self.get_orient_scores(vecs)
avg = np.mean(scores, axis=0)
avgvecs = np.mean(vecs, axis=0)
hist[label1][label2] = avg
orient_scores[:, ipair] = avg
#orient_prior[:,ipair] = avgvecs
#print self.label_pairs[ipair], avg
ipair += 1
#orient_scores /= np.c_[np.sum(orient_scores,axis=1)]
#import ipdb; ipdb.set_trace()
return orient_scores, hist
def load_or_compute_orientations(self,train,test, mask=None):
from rwsegment import boundary_utils
reload(boundary_utils)
## Train classifier
logger.info('train orientations with train {} for test {}'.format(train, test))
##load image and seg
seg = io_analyze.load(config.dir_reg + test + train + 'regseg.hdr').astype(int)
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## sample points
points = boundary_utils.sample_points(np.ones(seg.shape), 5, mask=mask)
logger.debug('number of sampled points = {}'.format(len(points)))
## find edges between muscles
points_label = seg[tuple(points.T)]
nlabel = len(self.labelset)
hist = {}
orient_scores = np.zeros((self.orients.shape[0],self.nlabel**2))
#orient_prior = np.ones((3,self.label_pairs.shape[0]))
ipair = 0
for l1 in range(nlabel):
label1 = self.labelset[l1]
inds1 = np.where(points_label==label1)[0]
hist[label1] = {}
for l2 in range(nlabel):
label2 = self.labelset[l2]
inds2 = np.where(points_label==label2)[0]
edges = np.argwhere(np.triu(np.ones((inds1.size, inds2.size)),k=1))
edges = np.c_[inds1[edges[:,0]], inds2[edges[:,1]]]
vecs = points[edges[:,1]] - points[edges[:,0]]
vecs = vecs / np.c_[np.sqrt(np.sum(vecs**2,axis=1))]
if l1==0 or l1==l2:
avg = np.ones(len(self.orients))/float(len(self.orients))
else:
scores = self.get_orient_scores(vecs)
avg = np.mean(scores,axis=0)
avgvecs = np.mean(vecs,axis=0)
hist[label1][label2] = avg
orient_scores[:,ipair] = avg
#orient_prior[:,ipair] = avgvecs
#print self.label_pairs[ipair], avg
ipair += 1
#orient_scores /= np.c_[np.sum(orient_scores,axis=1)]
#import ipdb; ipdb.set_trace()
return orient_scores,hist
def batch_compute_dice(folder_list, labelset=[13, 14, 15, 16]):
import config
dir_gt = config.dir_reg
lbset = np.asarray(labelset, dtype=int)
for folder in folder_list:
for vol in config.vols:
if not os.path.isfile(folder + '/' + vol + '/sol.hdr'):
continue
print 'computing dice for segmentation: {}'.format(folder + '/' +
vol +
'/sol.hdr')
sol = io_analyze.load(folder + '/' + vol + '/sol.hdr').astype(int)
gt = io_analyze.load(dir_gt + '/' + vol + '/seg.hdr').astype(int)
d_slice = compute_dice_per_slice(sol, gt, labelset=labelset)
d_labels = compute_dice_coef(sol, gt, labelset=labelset)
#print d_slice, d_labels
np.savetxt(folder + '/' + vol + '/dice_labels.txt',
np.c_[d_labels.keys(),
d_labels.values()],
fmt='%d %f')
np.savetxt(folder + '/' + vol + '/dice_slices.txt',
np.c_[d_slice.keys(), d_slice.values()],
fmt='%d %f')
def load_or_compute_prior_and_mask(test, force_recompute=False, pca=False, fold=None):
if fold is not None:
test_name = 'f{}_{}/'.format(fold[0][:2], test)
else:
test_name = test
fold = [test]
#import ipdb; ipdb.set_trace()
labelset = np.asarray(config.labelset)
if pca:
outdir = config.dir_pca_prior + test_name
else:
outdir = config.dir_prior + test_name
if not os.path.isdir(outdir):
os.makedirs(outdir)
## load mask and prior
prior = None
file_mask = outdir + 'mask.hdr'
file_prior = outdir + 'prior.npz'
file_U = outdir + 'eigenvectors.npy'
file_segprior = outdir + 'segprior.hdr'
file_entropymap = outdir + 'entropymap.hdr'
if force_recompute or not (os.path.exists(file_prior) and os.path.exists(file_mask)):
if pca:
_prior, mask = load_or_compute_prior_and_mask(test, fold=fold)
generator = rwsegment_pca_prior.PriorGenerator(labelset, mask=mask)
else:
generator = rwsegment_prior.PriorGenerator(labelset)
ntrain = 0
for train in config.vols:
if train in fold: continue
logger.debug('load training img: {}'.format(train))
## segmentation
file_seg = config.dir_reg + test + train + 'regseg.hdr'
seg = io_analyze.load(file_seg)
## image (for intensity prior)
file_im = config.dir_reg + test + train + 'reggray.hdr'
im = io_analyze.load(file_im)
generator.add_training_data(seg,image=im, nrandom=5)
ntrain += 1
if not pca:
from scipy import ndimage
mask = generator.get_mask()
struct = np.ones((7,)*mask.ndim)
mask = ndimage.binary_dilation(
mask.astype(bool),
structure=struct,
).astype(bool)
prior = generator.get_prior(mask)
#import ipdb; ipdb.set_trace()
nlabel = len(labelset)
segprior = np.zeros(mask.shape)
segprior.flat[prior['imask']] = labelset[np.argmax(prior['data'],axis=0)]
entropymap = np.zeros(mask.shape)
entropymap.flat[prior['imask']] = np.sum(
np.log(prior['data'] + 1e-10)*prior['data'],
axis=0)
entropymap = entropymap / np.log(nlabel) * 2**15
if 'eigenvectors' in prior:
U = prior.pop('eigenvectors')
print 'size of U {}, dtype={}'.format(U.size, U.dtype)
np.save(file_U, U)
prior['eigenvectors'] = U
np.savez(file_prior,**prior)
io_analyze.save(file_mask, mask.astype(np.int32))
io_analyze.save(file_segprior, segprior.astype(np.int32))
#io_analyze.save(file_entropymap, entropymap.astype(np.int32))
mask = io_analyze.load(file_mask).astype(bool)
prior = dict(np.load(file_prior))
if pca:
U = np.load(file_U)
prior['eigenvectors'] = U
return prior, mask
def process_sample(self,test,fold=None):
## get prior
prior, mask = load_or_compute_prior_and_mask(
test,
fold=fold,
force_recompute=self.force_recompute_prior)
seeds = (-1)*mask
## load image
file_name = config.dir_reg + test + 'gray.hdr'
logger.info('segmenting data: {}'.format(file_name))
im = io_analyze.load(file_name)
file_gt = config.dir_reg + test + 'seg.hdr'
seg = io_analyze.load(file_gt)
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## normalize image
nim = im/np.std(im)
## init anchor_api
anchor_api = MetaAnchor(
prior=prior,
prior_models=self.prior_models,
prior_weights=self.prior_weights,
image=nim,
)
## start segmenting
#import ipdb; ipdb.set_trace()
sol,y = rwsegment.segment(
nim,
anchor_api,
seeds=seeds,
labelset=self.labelset,
weight_function=self.weight_function,
**self.params
)
## compute losses
z = seg.ravel()==np.c_[self.labelset]
flatmask = mask.ravel()*np.ones((len(self.labelset),1))
## loss 0 : 1 - Dice(y,z)
loss0 = loss_functions.ideal_loss(z,y,mask=flatmask)
logger.info('Tloss = {}'.format(loss0))
## loss2: squared difference with ztilde
loss1 = loss_functions.anchor_loss(z,y,mask=flatmask)
logger.info('SDloss = {}'.format(loss1))
## loss3: laplacian loss
loss2 = loss_functions.laplacian_loss(z,y,mask=flatmask)
logger.info('LAPloss = {}'.format(loss2))
## loss4: linear loss
loss3 = loss_functions.linear_loss(z,y,mask=flatmask)
logger.info('LINloss = {}'.format(loss3))
## compute Dice coefficient per label
dice = compute_dice_coef(sol, seg,labelset=self.labelset)
logger.info('Dice: {}'.format(dice))
if not config.debug:
if fold is not None:
test_name = 'f{}_{}'.format(fold[0][:2], test)
else:
test_name = test
outdir = config.dir_seg + \
'/{}/{}'.format(self.model_name,test_name)
logger.info('saving data in: {}'.format(outdir))
if not os.path.isdir(outdir):
os.makedirs(outdir)
f = open(outdir + 'losses.txt', 'w')
f.write('ideal_loss\t{}\n'.format(loss0))
f.write('anchor_loss\t{}\n'.format(loss1))
f.write('laplacian_loss\t{}\n'.format(loss2))
f.close()
io_analyze.save(outdir + 'sol.hdr', sol.astype(np.int32))
np.savetxt(
outdir + 'dice.txt', np.c_[dice.keys(),dice.values()],fmt='%d %.8f')
def compute_mean_segmentation(self, list):
for test in list:
file_gt = config.dir_reg + test + 'seg.hdr'
seg = io_analyze.load(file_gt)
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## get prior
prior, mask = load_or_compute_prior_and_mask(
test,force_recompute=self.force_recompute_prior)
mask = mask.astype(bool)
y = np.zeros((len(self.labelset),seg.size))
y[:,0] = 1
y.flat[prior['imask']] = prior['data']
sol = np.zeros(seg.shape,dtype=np.int32)
sol[mask] = self.labelset[np.argmax(prior['data'],axis=0)]
## compute losses
z = seg.ravel()==np.c_[self.labelset]
flatmask = mask.ravel()*np.ones((len(self.labelset),1))
## loss 0 : 1 - Dice(y,z)
loss0 = loss_functions.ideal_loss(z,y,mask=flatmask)
logger.info('Tloss = {}'.format(loss0))
## loss2: squared difference with ztilde
#loss1 = loss_functions.anchor_loss(z,y,mask=flatmask)
#logger.info('SDloss = {}'.format(loss1))
## loss3: laplacian loss
#loss2 = loss_functions.laplacian_loss(z,y,mask=flatmask)
#logger.info('LAPloss = {}'.format(loss2))
## loss4: linear loss
#loss3 = loss_functions.linear_loss(z,y,mask=flatmask)
#logger.info('LINloss = {}'.format(loss3))
## compute Dice coefficient per label
dice = compute_dice_coef(sol, seg,labelset=self.labelset)
logger.info('Dice: {}'.format(dice))
if not config.debug:
outdir = config.dir_seg + \
'/{}/{}'.format('mean',test)
logger.info('saving data in: {}'.format(outdir))
if not os.path.isdir(outdir):
os.makedirs(outdir)
#f = open(outdir + 'losses.txt', 'w')
#f.write('ideal_loss\t{}\n'.format(loss0))
#f.write('anchor_loss\t{}\n'.format(loss1))
#f.write('laplacian_loss\t{}\n'.format(loss2))
#f.close()
io_analyze.save(outdir + 'sol.hdr', sol.astype(np.int32))
np.savetxt(
outdir + 'dice.txt', np.c_[dice.keys(),dice.values()],fmt='%d %.8f')
def make_training_set(self, test, fold=None):
if fold is None:
fold = [test]
## training images and segmentations
if self.isroot:
slice_border = 20 # do not consider top and bottom slices
images = []
segmentations = []
metadata = []
for train in self.training_vols:
if train in fold: continue
logger.info('loading training data: {}'.format(train))
## file names
file_seg = self.dir_reg + test + train + 'regseg.hdr'
file_im = self.dir_reg + test + train + 'reggray.hdr'
## load image
im = io_analyze.load(file_im)
im = im / np.std(im) # normalize image by std
## load segmentation
seg = io_analyze.load(file_seg).astype(int)
seg.flat[~np.in1d(seg.ravel(), self.labelset
)] = self.labelset[0]
if self.crop:
## if split training images into smaller sets
pmask = -1 * np.ones(seg.shape, dtype=int)
pmask.flat[self.prior['imask']] = np.arange(
len(self.prior['imask']))
nslice = im.shape[0]
for i in range(nslice / self.slice_step):
istart = i * self.slice_step
iend = np.minimum(
nslice, i * self.slice_step + self.slice_size)
if istart < slice_border or istart > (im.shape[0] -
slice_border):
continue
islices = np.arange(istart, iend)
if np.all(seg[islices]==self.labelset[0]) or \
np.all(self.seeds[islices]>=0):
continue
logger.debug(
'ivol {}, slices: start end: {} {}'.format(
len(images), istart, iend))
bin = (seg[islices].ravel() == np.c_[self.labelset]
) # make bin vector z
pmaski = pmask[islices]
imask = np.where(pmaski.ravel() > 0)[0]
iimask = pmaski.flat[imask]
#iimask = pmask[islices]
#iimask = iimask[iimask>=0]
## append to training set
images.append(im[islices])
segmentations.append(bin)
metadata.append({
'islices': islices,
'imask': imask,
'iimask': iimask
})
## break loop
if len(images) == self.select_vol.stop:
break
else:
bin = (seg.ravel() == np.c_[self.labelset]
) # make bin vector z
## append to training set
images.append(im)
segmentations.append(bin)
metadata.append({})
## break loop
if len(images) == self.select_vol.stop:
break
nmaxvol = 100
if len(images) > nmaxvol:
iselect = np.arange(len(images))
iselect = iselect[np.random.randint(
0,
len(iselect),
np.minimum(nmaxvol, len(iselect)),
)]
iselect = np.sort(iselect)
logger.info('selected training: {}'.format(iselect))
images = [images[i] for i in iselect]
segmentations = [segmentations[i] for i in iselect]
metadata = [metadata[i] for i in iselect]
ntrain = len(images)
logger.info('Learning with {} training examples'\
.format(ntrain))
self.training_set = (images, segmentations, metadata)
def run_svm_inference(self,test,w):
logger.info('running inference on: {}'.format(test))
outdir = self.dir_inf + test
if not os.path.isdir(outdir):
os.makedirs(outdir)
## segment test image with trained w
def wwf(im,_w):
''' meta weight function'''
data = 0
for iwf,wf in enumerate(self.weight_functions.values()):
ij,_data = wf(im)
data += _w[iwf]*_data
return ij, data
## load images and ground truth
file_seg = self.dir_reg + test + 'seg.hdr'
file_im = self.dir_reg + test + 'gray.hdr'
im = io_analyze.load(file_im)
seg = io_analyze.load(file_seg)
seg.flat[~np.in1d(seg.ravel(),self.labelset)] = self.labelset[0]
## save image
im = im/np.std(im) # normalize image by variance
## prior
anchor_api = BaseAnchorAPI(
self.prior,
anchor_weight=w[-1],
)
sol,y = rwsegment.segment(
im,
anchor_api,
seeds=self.seeds,
weight_function=lambda im: wwf(im, w),
**self.rwparams_inf
)
np.save(outdir + 'y.test.npy',y)
io_analyze.save(outdir + 'sol.test.hdr',sol.astype(np.int32))
## compute Dice coefficient
dice = compute_dice_coef(sol, seg,labelset=self.labelset)
np.savetxt(
outdir + 'dice.test.txt', np.c_[dice.keys(),dice.values()],fmt='%d %.8f')
## inference compare with gold standard
dice_gold = np.loadtxt(outdir + 'dice.gold.txt')
y_gold = np.load(outdir + 'y.gold.npy')
sol_gold = io_analyze.load(outdir + 'sol.gold.hdr')
np.testing.assert_allclose(
dice.values(),
dict(dice_gold).values(),
err_msg='FAIL: dice coef mismatch',
atol=1e-8)
np.testing.assert_allclose(y, y_gold, err_msg='FAIL: y mismatch')
np.testing.assert_equal(sol, sol_gold, err_msg='FAIL: sol mismatch')
print 'PASS: inference tests'
def process_sample(self, train, test):
outdir = config.dir_work + 'autoseeds/' + config.basis + '/' + train + '/' + test
logger.info('saving data in: {}'.format(outdir))
if not os.path.isdir(outdir):
os.makedirs(outdir)
## get prior
from scipy import ndimage
segtrain = io_analyze.load(config.dir_reg + test + train +
'/regseg.hdr')
segtrain.flat[~np.in1d(segtrain, self.labelset)] = self.labelset[0]
struct = np.ones((10, ) * segtrain.ndim)
mask = ndimage.binary_dilation(
segtrain > 0,
structure=struct,
).astype(bool)
#prior, mask = load_or_compute_prior_and_mask(
# test,force_recompute=self.force_recompute_prior)
#mask = mask.astype(bool)
## load image
file_name = config.dir_reg + test + 'gray.hdr'
logger.info('segmenting data: {}'.format(file_name))
im = io_analyze.load(file_name).astype(float)
file_gt = config.dir_reg + test + 'seg.hdr'
seg = io_analyze.load(file_gt)
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## normalize image
nim = im / np.std(im)
#orient_scores = self.load_or_compute_orientations(train,test, mask=mask)
if 1: #not os.path.isfile(outdir + 'points.npy'):
from rwsegment import boundary_utils
reload(boundary_utils)
## sample points
points = boundary_utils.sample_points(im, self.step, mask=mask)
points = points[mask[tuple(points.T)]]
impoints = np.zeros(im.shape, dtype=int)
impoints[tuple(points.T)] = np.arange(len(points)) + 1
ipoints = np.where(impoints.ravel())[0]
points = np.argwhere(impoints)
np.save(outdir + 'points.npy', points)
impoints[tuple(points.T)] = np.arange(len(points)) + 1
## set unary potentials from prior: array of unary costs
nlabel = len(self.labelset)
dist = self.distance_to_train(segtrain, points)
T = 10.0
prob_pts = np.exp(-(dist / T)**2) / np.c_[np.sum(
np.exp(-(dist / T)**2), axis=1)]
#prob = np.c_[np.ones(im.size), np.zeros((im.size, nlabel-1))]
#prob[mask.ravel(),:] = prior['data'].T
#prob_pts = prob[ipoints,:]
np.save(outdir + 'prob_points.npy', prob_pts)
## binary potentials
## compute edges
edges, edgev, labels = boundary_utils.get_edges(im,
points,
mask=mask)
edges = np.sort(edges, axis=1)
np.save(outdir + 'edges.npy', edges)
## get orientation hist
orient_scores, hist = self.load_or_compute_orientations(train,
test,
mask=mask)
##classify edges
vecs = points[edges[:, 1]] - points[edges[:, 0]]
vecs = vecs / np.c_[np.sqrt(np.sum(vecs**2, axis=1))]
scores = self.get_orient_scores(vecs)
prob_orient = np.dot(scores, orient_scores)
#prob_orient = prob_orient/np.c_[np.sum(prob_orient, axis=1)]
np.save(outdir + 'prob_orient.npy', prob_orient)
'''
## load classifier
classifier = self.load_or_compute_classifier(train,test, mask=mask)
## extract profiles
profiles,emap,dists = boundary_utils.get_profiles(nim, points, edges, rad=0)
## make features
x = boundary_utils.make_features(
profiles,
size=self.sizex,
additional=[dists,edgev,edgev/dists],
)
## classify
cl, scores = classifier.classify(x)
## ground truth
z = boundary_utils.is_boundary(points, edges, seg)
logger.info('non boundary classification: {}%'\
.format(np.sum((np.r_[z]==0)*(np.r_[cl]==0))/float(np.sum(np.r_[z]==0))*100))
logger.info('boundary classification: {}%'\
.format(np.sum((np.r_[z]==1)*(np.r_[cl]==1))/float(np.sum(np.r_[z]==1))*100))
np.save(outdir + 'classified.npy', cl)
## probabilities
prob_edges = 1. - scores/np.c_[np.sum(scores, axis=1)]
##save probs
np.save(outdir + 'prob_edges.npy',prob_edges)
'''
else:
points = np.load(outdir + 'points.npy')
edges = np.load(outdir + 'edges.npy')
cl = np.load(outdir + 'classified.npy')
prob_pts = np.load(outdir + 'prob_points.npy')
#prob_edges = np.load(outdir + 'prob_edges.npy')
prob_orient = np.load(outdir + 'prob_orient.npy')
## make potentials
unary = -np.log(prob_pts + 1e-10)
#binary = - np.log(prob_edges + 1e-10)
#thresh = (prob_orient.shape[1] - 1.0)/prob_orient.shape[1]
thresh = (len(self.orients) - 1.0) / len(self.orients)
orient_cost = -np.log(np.clip(prob_orient + thresh, 0, 1) +
1e-10) * 100
orient_cost = np.clip(orient_cost, 0, 1e10)
#import ipdb; ipdb.set_trace()
## solve MRF:
import ipdb
ipdb.set_trace()
'''
from rwsegment.mrf import fastPD
class CostFunction(object):
def __init__(self,**kwargs):
self.binary = kwargs.pop('binary',0)
self.orient_indices = kwargs.pop('orient_indices')
self.orient_cost = kwargs.pop('orient_cost')
def __call__(self,e,l1,l2):
idpair = self.orient_indices[l1,l2]
pair_cost = self.orient_cost[e,idpair]
cost = (l1!=l2)*pair_cost
#return (l1!=l2)*(1-cl[e])*0.1
#return (l1!=l2)*self.binary[e,1]*0.1
#y = l1!=l2
#return self.binary[e, y]*pair_cost
print e, l1, l2, cost
return cost
#sol, en = fastPD.fastPD_callback(unary, edges, cost_function(binary), debug=True)
cost_function = CostFunction(
#binary=binary,
orient_indices=self.orient_indices,
orient_cost=orient_cost,
)
sol, en = fastPD.fastPD_callback(unary, edges, cost_function, debug=True)
'''
wpairs = orient_cost
from rwsegment.mrf import trw
sol, en = trw.TRW_general(unary,
edges,
wpairs,
niters=1000,
verbose=True)
labels = self.labelset[sol]
imsol = np.ones(im.shape, dtype=np.int32) * 20
imsol[tuple(points.T)] = labels
io_analyze.save(outdir + 'imseeds.hdr', imsol)
## classify sol
gtlabels = seg[tuple(points.T)]
priorlabels = self.labelset[np.argmin(unary, axis=1)]
err_prior = 1 - np.sum(gtlabels == priorlabels) / float(len(points))
err = 1 - np.sum(gtlabels == labels) / float(len(points))
logger.info('error in prior sol: {}%'.format(err_prior * 100))
logger.info('error in sol: {}%'.format(err * 100))
import ipdb
ipdb.set_trace()
## start segmenting
sol, y = rwsegment.segment(nim,
seeds=seeds,
labelset=self.labelset,
weight_function=self.weight_function,
**self.params)
## compute Dice coefficient per label
dice = compute_dice_coef(sol, seg, labelset=self.labelset)
logger.info('Dice: {}'.format(dice))
if not config.debug:
io_analyze.save(outdir + 'sol.hdr', sol.astype(np.int32))
np.savetxt(outdir + 'dice.txt',
np.c_[dice.keys(), dice.values()],
fmt='%d %.8f')
def process_sample(self,train, test):
outdir = config.dir_work + 'autoseeds/' + config.basis + '/' + train + '/' + test
logger.info('saving data in: {}'.format(outdir))
if not os.path.isdir(outdir):
os.makedirs(outdir)
## get prior
from scipy import ndimage
segtrain = io_analyze.load(config.dir_reg + test + train + '/regseg.hdr')
segtrain.flat[~np.in1d(segtrain, self.labelset)] = self.labelset[0]
struct = np.ones((10,)*segtrain.ndim)
mask = ndimage.binary_dilation(
segtrain>0,
structure=struct,
).astype(bool)
#prior, mask = load_or_compute_prior_and_mask(
# test,force_recompute=self.force_recompute_prior)
#mask = mask.astype(bool)
## load image
file_name = config.dir_reg + test + 'gray.hdr'
logger.info('segmenting data: {}'.format(file_name))
im = io_analyze.load(file_name).astype(float)
file_gt = config.dir_reg + test + 'seg.hdr'
seg = io_analyze.load(file_gt)
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## normalize image
nim = im/np.std(im)
#orient_scores = self.load_or_compute_orientations(train,test, mask=mask)
if 1:#not os.path.isfile(outdir + 'points.npy'):
from rwsegment import boundary_utils
reload(boundary_utils)
## sample points
points = boundary_utils.sample_points(im, self.step, mask=mask)
points = points[mask[tuple(points.T)]]
impoints = np.zeros(im.shape,dtype=int)
impoints[tuple(points.T)] = np.arange(len(points)) + 1
ipoints = np.where(impoints.ravel())[0]
points = np.argwhere(impoints)
np.save(outdir + 'points.npy', points)
impoints[tuple(points.T)] = np.arange(len(points)) + 1
## set unary potentials from prior: array of unary costs
nlabel = len(self.labelset)
dist = self.distance_to_train(segtrain, points)
T = 10.0
prob_pts = np.exp(-(dist/T)**2) / np.c_[np.sum(np.exp(-(dist/T)**2),axis=1)]
#prob = np.c_[np.ones(im.size), np.zeros((im.size, nlabel-1))]
#prob[mask.ravel(),:] = prior['data'].T
#prob_pts = prob[ipoints,:]
np.save(outdir + 'prob_points.npy', prob_pts)
## binary potentials
## compute edges
edges,edgev,labels = boundary_utils.get_edges(im, points, mask=mask)
edges = np.sort(edges,axis=1)
np.save(outdir + 'edges.npy', edges)
## get orientation hist
orient_scores,hist = self.load_or_compute_orientations(train,test, mask=mask)
##classify edges
vecs = points[edges[:,1]] - points[edges[:,0]]
vecs = vecs / np.c_[np.sqrt(np.sum(vecs**2,axis=1))]
scores = self.get_orient_scores(vecs)
prob_orient = np.dot(scores, orient_scores)
#prob_orient = prob_orient/np.c_[np.sum(prob_orient, axis=1)]
np.save(outdir + 'prob_orient.npy', prob_orient)
'''
## load classifier
classifier = self.load_or_compute_classifier(train,test, mask=mask)
## extract profiles
profiles,emap,dists = boundary_utils.get_profiles(nim, points, edges, rad=0)
## make features
x = boundary_utils.make_features(
profiles,
size=self.sizex,
additional=[dists,edgev,edgev/dists],
)
## classify
cl, scores = classifier.classify(x)
## ground truth
z = boundary_utils.is_boundary(points, edges, seg)
logger.info('non boundary classification: {}%'\
.format(np.sum((np.r_[z]==0)*(np.r_[cl]==0))/float(np.sum(np.r_[z]==0))*100))
logger.info('boundary classification: {}%'\
.format(np.sum((np.r_[z]==1)*(np.r_[cl]==1))/float(np.sum(np.r_[z]==1))*100))
np.save(outdir + 'classified.npy', cl)
## probabilities
prob_edges = 1. - scores/np.c_[np.sum(scores, axis=1)]
##save probs
np.save(outdir + 'prob_edges.npy',prob_edges)
'''
else:
points = np.load(outdir + 'points.npy')
edges = np.load(outdir + 'edges.npy')
cl = np.load(outdir + 'classified.npy')
prob_pts = np.load(outdir + 'prob_points.npy')
#prob_edges = np.load(outdir + 'prob_edges.npy')
prob_orient = np.load(outdir + 'prob_orient.npy')
## make potentials
unary = - np.log(prob_pts + 1e-10)
#binary = - np.log(prob_edges + 1e-10)
#thresh = (prob_orient.shape[1] - 1.0)/prob_orient.shape[1]
thresh = (len(self.orients) - 1.0) / len(self.orients)
orient_cost = - np.log(np.clip(prob_orient + thresh,0,1) + 1e-10)*100
orient_cost = np.clip(orient_cost, 0, 1e10)
#import ipdb; ipdb.set_trace()
## solve MRF:
import ipdb; ipdb.set_trace()
'''
from rwsegment.mrf import fastPD
class CostFunction(object):
def __init__(self,**kwargs):
self.binary = kwargs.pop('binary',0)
self.orient_indices = kwargs.pop('orient_indices')
self.orient_cost = kwargs.pop('orient_cost')
def __call__(self,e,l1,l2):
idpair = self.orient_indices[l1,l2]
pair_cost = self.orient_cost[e,idpair]
cost = (l1!=l2)*pair_cost
#return (l1!=l2)*(1-cl[e])*0.1
#return (l1!=l2)*self.binary[e,1]*0.1
#y = l1!=l2
#return self.binary[e, y]*pair_cost
print e, l1, l2, cost
return cost
#sol, en = fastPD.fastPD_callback(unary, edges, cost_function(binary), debug=True)
cost_function = CostFunction(
#binary=binary,
orient_indices=self.orient_indices,
orient_cost=orient_cost,
)
sol, en = fastPD.fastPD_callback(unary, edges, cost_function, debug=True)
'''
wpairs = orient_cost
from rwsegment.mrf import trw
sol, en = trw.TRW_general(
unary, edges, wpairs, niters=1000, verbose=True)
labels = self.labelset[sol]
imsol = np.ones(im.shape, dtype=np.int32)*20
imsol[tuple(points.T)] = labels
io_analyze.save(outdir + 'imseeds.hdr', imsol)
## classify sol
gtlabels = seg[tuple(points.T)]
priorlabels = self.labelset[np.argmin(unary,axis=1)]
err_prior = 1 - np.sum(gtlabels==priorlabels)/float(len(points))
err = 1 - np.sum(gtlabels==labels)/float(len(points))
logger.info('error in prior sol: {}%'.format(err_prior*100))
logger.info('error in sol: {}%'.format(err*100))
import ipdb; ipdb.set_trace()
## start segmenting
sol,y = rwsegment.segment(
nim,
seeds=seeds,
labelset=self.labelset,
weight_function=self.weight_function,
**self.params
)
## compute Dice coefficient per label
dice = compute_dice_coef(sol, seg,labelset=self.labelset)
logger.info('Dice: {}'.format(dice))
if not config.debug:
io_analyze.save(outdir + 'sol.hdr', sol.astype(np.int32))
np.savetxt(
outdir + 'dice.txt', np.c_[dice.keys(),dice.values()],fmt='%d %.8f')
def run_svm_inference(self, test, w, test_dir):
logger.info('running inference on: {}'.format(test))
## normalize w
# w = w / np.sqrt(np.dot(w,w))
strw = ' '.join('{:.3}'.format(val)
for val in np.asarray(w) * self.psi_scale)
logger.debug('scaled w=[{}]'.format(strw))
weights_laplacians = np.asarray(w)[self.indices_laplacians]
weights_laplacians_h = np.asarray(
self.hand_tuned_w)[self.indices_laplacians]
weights_priors = np.asarray(w)[self.indices_priors]
weights_priors_h = np.asarray(self.hand_tuned_w)[self.indices_priors]
## segment test image with trained w
'''
def meta_weight_functions(im,i,j,_w):
data = 0
for iwf,wf in enumerate(self.laplacian_functions):
_data = wf(im,i,j)
data += _w[iwf]*_data
return data
weight_function = lambda im: meta_weight_functions(im,i,j,weights_laplacians)
weight_function_h = lambda im: meta_weight_functions(im,i,j,weights_laplacians_h)
'''
weight_function = MetaLaplacianFunction(weights_laplacians,
self.laplacian_functions)
weight_function_h = MetaLaplacianFunction(weights_laplacians_h,
self.laplacian_functions)
## load images and ground truth
file_seg = self.dir_reg + test + 'seg.hdr'
file_im = self.dir_reg + test + 'gray.hdr'
im = io_analyze.load(file_im)
seg = io_analyze.load(file_seg)
seg.flat[~np.in1d(seg.ravel(), self.labelset)] = self.labelset[0]
nim = im / np.std(im) # normalize image by std
## test training data ?
inference_train = True
if inference_train:
train_ims, train_segs, train_metas = self.training_set
for tim, tz, tmeta in zip(train_ims, train_segs, train_metas):
## retrieve metadata
islices = tmeta.pop('islices', None)
imask = tmeta.pop('imask', None)
iimask = tmeta.pop('iimask', None)
if islices is not None:
tseeds = self.seeds[islices]
tprior = {
'data': np.asarray(self.prior['data'])[:, iimask],
'imask': imask,
'variance': np.asarray(self.prior['variance'])[:,
iimask],
'labelset': self.labelset,
}
if 'intensity' in self.prior:
tprior['intensity'] = self.prior['intensity']
else:
tseeds = self.seeds
tprior = self.prior
## prior
tseg = self.labelset[np.argmax(tz, axis=0)].reshape(tim.shape)
tanchor_api = MetaAnchor(
tprior,
self.prior_functions,
weights_priors,
image=tim,
)
tsol, ty = rwsegment.segment(tim,
tanchor_api,
seeds=tseeds,
weight_function=weight_function,
**self.rwparams_inf)
## compute Dice coefficient
tdice = compute_dice_coef(tsol, tseg, labelset=self.labelset)
logger.info('Dice coefficients for train: \n{}'.format(tdice))
nlabel = len(self.labelset)
tflatmask = np.zeros(ty.shape, dtype=bool)
tflatmask[:, imask] = True
loss0 = loss_functions.ideal_loss(tz, ty, mask=tflatmask)
logger.info('Tloss = {}'.format(loss0))
## loss2: squared difference with ztilde
loss1 = loss_functions.anchor_loss(tz, ty, mask=tflatmask)
logger.info('SDloss = {}'.format(loss1))
## loss3: laplacian loss
loss2 = loss_functions.laplacian_loss(tz, ty, mask=tflatmask)
logger.info('LAPloss = {}'.format(loss2))
tanchor_api_h = MetaAnchor(
tprior,
self.prior_functions,
weights_priors_h,
image=tim,
)
tsol, ty = rwsegment.segment(tim,
tanchor_api_h,
seeds=tseeds,
weight_function=weight_function_h,
**self.rwparams_inf)
## compute Dice coefficient
tdice = compute_dice_coef(tsol, tseg, labelset=self.labelset)
logger.info(
'Dice coefficients for train (hand-tuned): \n{}'.format(
tdice))
loss0 = loss_functions.ideal_loss(tz, ty, mask=tflatmask)
logger.info('Tloss (hand-tuned) = {}'.format(loss0))
## loss2: squared difference with ztilde
loss1 = loss_functions.anchor_loss(tz, ty, mask=tflatmask)
logger.info('SDloss (hand-tuned) = {}'.format(loss1))
## loss3: laplacian loss
loss2 = loss_functions.laplacian_loss(tz, ty, mask=tflatmask)
logger.info('LAPloss (hand-tuned) = {}'.format(loss2))
break
## prior
anchor_api = MetaAnchor(
self.prior,
self.prior_functions,
weights_priors,
image=nim,
)
sol, y = rwsegment.segment(nim,
anchor_api,
seeds=self.seeds,
weight_function=weight_function,
**self.rwparams_inf)
## compute Dice coefficient
dice = compute_dice_coef(sol, seg, labelset=self.labelset)
logger.info('Dice coefficients: \n{}'.format(dice))
## objective
en_rw = rwsegment.energy_rw(nim,
y,
seeds=self.seeds,
weight_function=weight_function,
**self.rwparams_inf)
en_anchor = rwsegment.energy_anchor(nim,
y,
anchor_api,
seeds=self.seeds,
**self.rwparams_inf)
obj = en_rw + en_anchor
logger.info('Objective = {:.3}'.format(obj))
## compute losses
z = seg.ravel() == np.c_[self.labelset]
mask = self.seeds < 0
flatmask = mask.ravel() * np.ones((len(self.labelset), 1))
## loss 0 : 1 - Dice(y,z)
loss0 = loss_functions.ideal_loss(z, y, mask=flatmask)
logger.info('Tloss = {}'.format(loss0))
## loss2: squared difference with ztilde
loss1 = loss_functions.anchor_loss(z, y, mask=flatmask)
logger.info('SDloss = {}'.format(loss1))
## loss3: laplacian loss
loss2 = loss_functions.laplacian_loss(z, y, mask=flatmask)
logger.info('LAPloss = {}'.format(loss2))
## loss4: linear loss
loss3 = loss_functions.linear_loss(z, y, mask=flatmask)
logger.info('LINloss = {}'.format(loss3))
## saving
if self.debug:
pass
elif self.isroot:
outdir = self.dir_inf + test_dir
logger.info('saving data in: {}'.format(outdir))
if not os.path.isdir(outdir):
os.makedirs(outdir)
#io_analyze.save(outdir + 'im.hdr',im.astype(np.int32))
#np.save(outdir + 'y.npy',y)
#io_analyze.save(outdir + 'sol.hdr',sol.astype(np.int32))
np.savetxt(outdir + 'objective.txt', [obj])
np.savetxt(outdir + 'dice.txt',
np.c_[dice.keys(), dice.values()],
fmt='%d %f')
f = open(outdir + 'losses.txt', 'w')
f.write('ideal_loss\t{}\n'.format(loss0))
f.write('anchor_loss\t{}\n'.format(loss1))
f.write('laplacian_loss\t{}\n'.format(loss2))
f.close()
def compute_objective(test, y, w):
im = io_analyze.load(config.dir_reg + test + 'gray.hdr')
nim = im / np.std(im)
prior, mask = load_or_compute_prior_and_mask(test, force_recompute=False)
seeds = (-1) * mask.astype(int)
from rwsegment import rwsegment_prior_models as models
from rwsegment import weight_functions as wflib
rwparams = {
'labelset': np.asarray(config.labelset),
# optimization
'rtol': 1e-6,
'maxiter': 1e3,
'per_label': True,
'optim_solver': 'unconstrained',
}
weight_functions = {
'std_b10': lambda im: wflib.weight_std(im, beta=10),
'std_b50': lambda im: wflib.weight_std(im, beta=50),
'std_b100': lambda im: wflib.weight_std(im, beta=100),
'inv_b100o1': lambda im: wflib.weight_inv(im, beta=100, offset=1),
# 'pdiff_r1b10': lambda im: wflib.weight_patch_diff(im, r0=1, beta=10),
# 'pdiff_r2b10': lambda im: wflib.weight_patch_diff(im, r0=2, beta=10),
# 'pdiff_r1b50' : lambda im: wflib.weight_patch_diff(im, r0=1, beta=50),
}
prior_models = {
'constant': models.Constant,
'entropy': models.Entropy_no_D,
'intensity': models.Intensity,
}
## indices of w
nlaplacian = len(weight_functions)
nprior = len(prior_models)
indices_laplacians = np.arange(nlaplacian)
indices_priors = np.arange(nlaplacian, nlaplacian + nprior)
laplacian_functions = weight_functions.values()
laplacian_names = weight_functions.keys()
prior_functions = prior_models.values()
prior_names = prior_models.keys()
weights_laplacians = np.asarray(w)[indices_laplacians]
weights_priors = np.asarray(w)[indices_priors]
def meta_weight_functions(im, _w):
''' meta weight function'''
data = 0
for iwf, wf in enumerate(laplacian_functions):
ij, _data = wf(im)
data += _w[iwf] * _data
return ij, data
weight_function = lambda im: meta_weight_functions(im, weights_laplacians)
from svm_rw_api import MetaAnchor
anchor_api = MetaAnchor(
prior,
prior_functions,
weights_priors,
image=nim,
)
from rwsegment import rwsegment
en_rw = rwsegment.energy_rw(nim,
y,
seeds=seeds,
weight_function=weight_function,
**rwparams)
en_anchor = rwsegment.energy_anchor(nim,
y,
anchor_api,
seeds=seeds,
**rwparams)
obj = en_rw + en_anchor
return obj
def compute_features(test, train, y):
im = io_analyze.load(config.dir_reg + test + train + 'reggray.hdr')
nim = im / np.std(im)
prior, mask = load_or_compute_prior_and_mask(test, force_recompute=False)
seeds = (-1) * mask.astype(int)
from rwsegment import rwsegment_prior_models as models
from rwsegment import weight_functions as wflib
rwparams = {
'labelset': np.asarray(config.labelset),
}
weight_functions = {
'std_b10': lambda im: wflib.weight_std(im, beta=10),
'std_b50': lambda im: wflib.weight_std(im, beta=50),
'std_b100': lambda im: wflib.weight_std(im, beta=100),
'inv_b100o1': lambda im: wflib.weight_inv(im, beta=100, offset=1),
# 'pdiff_r1b10': lambda im: wflib.weight_patch_diff(im, r0=1, beta=10),
# 'pdiff_r2b10': lambda im: wflib.weight_patch_diff(im, r0=2, beta=10),
# 'pdiff_r1b50' : lambda im: wflib.weight_patch_diff(im, r0=1, beta=50),
}
prior_models = {
'constant': models.Constant,
'entropy': models.Entropy_no_D,
'intensity': models.Intensity,
}
## indices of w
nlaplacian = len(weight_functions)
nprior = len(prior_models)
indices_laplacians = np.arange(nlaplacian)
indices_priors = np.arange(nlaplacian, nlaplacian + nprior)
laplacian_functions = weight_functions.values()
laplacian_names = weight_functions.keys()
prior_functions = prior_models.values()
prior_names = prior_models.keys()
#from svm_rw_api import MetaAnchor
#anchor_api = MetaAnchor(
# prior,
# prior_functions,
# weights_priors,
# image=im,
# )
from rwsegment import rwsegment
for fname in weight_functions:
wf = weight_functions[fname]
en_rw = rwsegment.energy_rw(nim,
y,
seeds=seeds,
weight_function=wf,
**rwparams)
print fname, en_rw
for mname in prior_models:
pm = prior_models[mname](prior, 1., image=nim)
en_anchor = rwsegment.energy_anchor(nim,
y,
pm,
seeds=seeds,
**rwparams)
print mname, en_anchor
def load_or_compute_prior_and_mask(test,
force_recompute=False,
pca=False,
fold=None):
if fold is not None:
test_name = 'f{}_{}/'.format(fold[0][:2], test)
else:
test_name = test
fold = [test]
#import ipdb; ipdb.set_trace()
labelset = np.asarray(config.labelset)
if pca:
outdir = config.dir_pca_prior + test_name
else:
outdir = config.dir_prior + test_name
if not os.path.isdir(outdir):
os.makedirs(outdir)
## load mask and prior
prior = None
file_mask = outdir + 'mask.hdr'
file_prior = outdir + 'prior.npz'
file_U = outdir + 'eigenvectors.npy'
file_segprior = outdir + 'segprior.hdr'
file_entropymap = outdir + 'entropymap.hdr'
if force_recompute or not (os.path.exists(file_prior)
and os.path.exists(file_mask)):
if pca:
_prior, mask = load_or_compute_prior_and_mask(test, fold=fold)
generator = rwsegment_pca_prior.PriorGenerator(labelset, mask=mask)
else:
generator = rwsegment_prior.PriorGenerator(labelset)
ntrain = 0
for train in config.vols:
if train in fold: continue
logger.debug('load training img: {}'.format(train))
## segmentation
file_seg = config.dir_reg + test + train + 'regseg.hdr'
seg = io_analyze.load(file_seg)
## image (for intensity prior)
file_im = config.dir_reg + test + train + 'reggray.hdr'
im = io_analyze.load(file_im)
generator.add_training_data(seg, image=im, nrandom=5)
ntrain += 1
if not pca:
from scipy import ndimage
mask = generator.get_mask()
struct = np.ones((7, ) * mask.ndim)
mask = ndimage.binary_dilation(
mask.astype(bool),
structure=struct,
).astype(bool)
prior = generator.get_prior(mask)
#import ipdb; ipdb.set_trace()
nlabel = len(labelset)
segprior = np.zeros(mask.shape)
segprior.flat[prior['imask']] = labelset[np.argmax(prior['data'],
axis=0)]
entropymap = np.zeros(mask.shape)
entropymap.flat[prior['imask']] = np.sum(
np.log(prior['data'] + 1e-10) * prior['data'], axis=0)
entropymap = entropymap / np.log(nlabel) * 2**15
if 'eigenvectors' in prior:
U = prior.pop('eigenvectors')
print 'size of U {}, dtype={}'.format(U.size, U.dtype)
np.save(file_U, U)
prior['eigenvectors'] = U
np.savez(file_prior, **prior)
io_analyze.save(file_mask, mask.astype(np.int32))
io_analyze.save(file_segprior, segprior.astype(np.int32))
#io_analyze.save(file_entropymap, entropymap.astype(np.int32))
mask = io_analyze.load(file_mask).astype(bool)
prior = dict(np.load(file_prior))
if pca:
U = np.load(file_U)
prior['eigenvectors'] = U
return prior, mask
def make_training_set(self,test, fold=None):
if fold is None:
fold = [test]
## training images and segmentations
if self.isroot:
slice_border = 20 # do not consider top and bottom slices
images = []
segmentations = []
metadata = []
for train in self.training_vols:
if train in fold: continue
logger.info('loading training data: {}'.format(train))
## file names
file_seg = self.dir_reg + test + train + 'regseg.hdr'
file_im = self.dir_reg + test + train + 'reggray.hdr'
## load image
im = io_analyze.load(file_im)
im = im/np.std(im) # normalize image by std
## load segmentation
seg = io_analyze.load(file_seg).astype(int)
seg.flat[~np.in1d(seg.ravel(),self.labelset)] = self.labelset[0]
if self.crop:
## if split training images into smaller sets
pmask = -1 * np.ones(seg.shape, dtype=int)
pmask.flat[self.prior['imask']] = np.arange(len(self.prior['imask']))
nslice = im.shape[0]
for i in range(nslice/self.slice_step):
istart = i*self.slice_step
iend = np.minimum(nslice, i*self.slice_step + self.slice_size)
if istart < slice_border or istart > (im.shape[0] - slice_border):
continue
islices = np.arange(istart, iend)
if np.all(seg[islices]==self.labelset[0]) or \
np.all(self.seeds[islices]>=0):
continue
logger.debug('ivol {}, slices: start end: {} {}'.format(len(images),istart, iend))
bin = (seg[islices].ravel()==np.c_[self.labelset]) # make bin vector z
pmaski = pmask[islices]
imask = np.where(pmaski.ravel()>0)[0]
iimask = pmaski.flat[imask]
#iimask = pmask[islices]
#iimask = iimask[iimask>=0]
## append to training set
images.append(im[islices])
segmentations.append(bin)
metadata.append({'islices': islices, 'imask':imask , 'iimask': iimask})
## break loop
if len(images) == self.select_vol.stop:
break
else:
bin = (seg.ravel()==np.c_[self.labelset])# make bin vector z
## append to training set
images.append(im)
segmentations.append(bin)
metadata.append({})
## break loop
if len(images) == self.select_vol.stop:
break
nmaxvol = 100
if len(images) > nmaxvol:
iselect = np.arange(len(images))
iselect = iselect[np.random.randint(
0,len(iselect),
np.minimum(nmaxvol, len(iselect)),
)]
iselect = np.sort(iselect)
logger.info('selected training: {}'.format(iselect))
images = [images[i] for i in iselect]
segmentations = [segmentations[i] for i in iselect]
metadata = [metadata[i] for i in iselect]
ntrain = len(images)
logger.info('Learning with {} training examples'\
.format(ntrain))
self.training_set = (images, segmentations, metadata)
def compute_objective(test, y, w):
im = io_analyze.load(config.dir_reg + test + 'gray.hdr')
nim = im/np.std(im)
prior,mask = load_or_compute_prior_and_mask(
test, force_recompute=False)
seeds = (-1)*mask.astype(int)
from rwsegment import rwsegment_prior_models as models
from rwsegment import weight_functions as wflib
rwparams = {
'labelset': np.asarray(config.labelset),
# optimization
'rtol': 1e-6,
'maxiter': 1e3,
'per_label':True,
'optim_solver':'unconstrained',
}
weight_functions = {
'std_b10' : lambda im: wflib.weight_std(im, beta=10),
'std_b50' : lambda im: wflib.weight_std(im, beta=50),
'std_b100' : lambda im: wflib.weight_std(im, beta=100),
'inv_b100o1' : lambda im: wflib.weight_inv(im, beta=100, offset=1),
# 'pdiff_r1b10': lambda im: wflib.weight_patch_diff(im, r0=1, beta=10),
# 'pdiff_r2b10': lambda im: wflib.weight_patch_diff(im, r0=2, beta=10),
# 'pdiff_r1b50' : lambda im: wflib.weight_patch_diff(im, r0=1, beta=50),
}
prior_models = {
'constant': models.Constant,
'entropy': models.Entropy_no_D,
'intensity': models.Intensity,
}
## indices of w
nlaplacian = len(weight_functions)
nprior = len(prior_models)
indices_laplacians = np.arange(nlaplacian)
indices_priors = np.arange(nlaplacian,nlaplacian + nprior)
laplacian_functions = weight_functions.values()
laplacian_names = weight_functions.keys()
prior_functions = prior_models.values()
prior_names = prior_models.keys()
weights_laplacians = np.asarray(w)[indices_laplacians]
weights_priors = np.asarray(w)[indices_priors]
def meta_weight_functions(im,_w):
''' meta weight function'''
data = 0
for iwf,wf in enumerate(laplacian_functions):
ij,_data = wf(im)
data += _w[iwf]*_data
return ij, data
weight_function = lambda im: meta_weight_functions(im, weights_laplacians)
from svm_rw_api import MetaAnchor
anchor_api = MetaAnchor(
prior,
prior_functions,
weights_priors,
image=nim,
)
from rwsegment import rwsegment
en_rw = rwsegment.energy_rw(
nim,
y,
seeds=seeds,
weight_function=weight_function,
**rwparams
)
en_anchor = rwsegment.energy_anchor(
nim,
y,
anchor_api,
seeds=seeds,
**rwparams
)
obj = en_rw + en_anchor
return obj
use_gt = True
if '--method' in sys.argv:
i = sys.argv.index('--method')
method = sys.argv[i + 1]
if '--slice' in sys.argv:
i = sys.argv.index('--slice')
islice = int(sys.argv[i + 1])
if '--test' in sys.argv:
i = sys.argv.index('--test')
test = sys.argv[i + 1] + '/'
print sys.argv
from rwsegment import io_analyze
dir = methods[method][2] + methods[method][1] + '/'
print dir
print test
print islice
sol = io_analyze.load(dir + test + 'sol.hdr')[islice]
seg = io_analyze.load(config.dir_reg + test + 'seg.hdr')[islice]
gray = io_analyze.load(config.dir_reg + test + 'gray.hdr')[islice]
if use_gt:
method = 'gt'
plot_cross(gray, seg, seg)
else:
plot_cross(gray, sol, seg)
pyplot.savefig('/home/baudinpy/plots/cross_{}_{}_{}.png'.format(
method, test[:-1], islice))
print 'saving to: {}'.format(
'/home/baudinpy/plots/cross_{}_{}_{}.png'.format(
method, test[:-1], islice))
def compute_features(test, train, y):
im = io_analyze.load(config.dir_reg + test + train + 'reggray.hdr')
nim = im/np.std(im)
prior,mask = load_or_compute_prior_and_mask(
test, force_recompute=False)
seeds = (-1)*mask.astype(int)
from rwsegment import rwsegment_prior_models as models
from rwsegment import weight_functions as wflib
rwparams = {
'labelset': np.asarray(config.labelset),
}
weight_functions = {
'std_b10' : lambda im: wflib.weight_std(im, beta=10),
'std_b50' : lambda im: wflib.weight_std(im, beta=50),
'std_b100' : lambda im: wflib.weight_std(im, beta=100),
'inv_b100o1' : lambda im: wflib.weight_inv(im, beta=100, offset=1),
# 'pdiff_r1b10': lambda im: wflib.weight_patch_diff(im, r0=1, beta=10),
# 'pdiff_r2b10': lambda im: wflib.weight_patch_diff(im, r0=2, beta=10),
# 'pdiff_r1b50' : lambda im: wflib.weight_patch_diff(im, r0=1, beta=50),
}
prior_models = {
'constant': models.Constant,
'entropy': models.Entropy_no_D,
'intensity': models.Intensity,
}
## indices of w
nlaplacian = len(weight_functions)
nprior = len(prior_models)
indices_laplacians = np.arange(nlaplacian)
indices_priors = np.arange(nlaplacian,nlaplacian + nprior)
laplacian_functions = weight_functions.values()
laplacian_names = weight_functions.keys()
prior_functions = prior_models.values()
prior_names = prior_models.keys()
#from svm_rw_api import MetaAnchor
#anchor_api = MetaAnchor(
# prior,
# prior_functions,
# weights_priors,
# image=im,
# )
from rwsegment import rwsegment
for fname in weight_functions:
wf = weight_functions[fname]
en_rw = rwsegment.energy_rw(
nim,
y,
seeds=seeds,
weight_function=wf,
**rwparams
)
print fname, en_rw
for mname in prior_models:
pm = prior_models[mname](prior,1., image=nim)
en_anchor = rwsegment.energy_anchor(
nim,
y,
pm,
seeds=seeds,
**rwparams
)
print mname, en_anchor
if '--gt' in sys.argv:
use_gt = True
if '--method' in sys.argv:
i = sys.argv.index('--method')
method = sys.argv[i+1]
if '--slice' in sys.argv:
i = sys.argv.index('--slice')
islice = int(sys.argv[i+1])
if '--test' in sys.argv:
i = sys.argv.index('--test')
test = sys.argv[i+1] + '/'
print sys.argv
from rwsegment import io_analyze
dir = methods[method][2] + methods[method][1] + '/'
print dir
print test
print islice
sol = io_analyze.load(dir + test + 'sol.hdr')[islice]
seg = io_analyze.load(config.dir_reg + test + 'seg.hdr')[islice]
gray = io_analyze.load(config.dir_reg + test + 'gray.hdr')[islice]
if use_gt:
method = 'gt'
plot_cross(gray, seg, seg)
else:
plot_cross(gray, sol, seg)
pyplot.savefig('/home/baudinpy/plots/cross_{}_{}_{}.png'.format(method, test[:-1], islice))
print 'saving to: {}'.format('/home/baudinpy/plots/cross_{}_{}_{}.png'.format(method, test[:-1], islice))
def run_svm_inference(self,test,w, test_dir):
logger.info('running inference on: {}'.format(test))
## normalize w
# w = w / np.sqrt(np.dot(w,w))
strw = ' '.join('{:.3}'.format(val) for val in np.asarray(w)*self.psi_scale)
logger.debug('scaled w=[{}]'.format(strw))
weights_laplacians = np.asarray(w)[self.indices_laplacians]
weights_laplacians_h = np.asarray(self.hand_tuned_w)[self.indices_laplacians]
weights_priors = np.asarray(w)[self.indices_priors]
weights_priors_h = np.asarray(self.hand_tuned_w)[self.indices_priors]
## segment test image with trained w
'''
def meta_weight_functions(im,i,j,_w):
data = 0
for iwf,wf in enumerate(self.laplacian_functions):
_data = wf(im,i,j)
data += _w[iwf]*_data
return data
weight_function = lambda im: meta_weight_functions(im,i,j,weights_laplacians)
weight_function_h = lambda im: meta_weight_functions(im,i,j,weights_laplacians_h)
'''
weight_function = MetaLaplacianFunction(
weights_laplacians,
self.laplacian_functions)
weight_function_h = MetaLaplacianFunction(
weights_laplacians_h,
self.laplacian_functions)
## load images and ground truth
file_seg = self.dir_reg + test + 'seg.hdr'
file_im = self.dir_reg + test + 'gray.hdr'
im = io_analyze.load(file_im)
seg = io_analyze.load(file_seg)
seg.flat[~np.in1d(seg.ravel(),self.labelset)] = self.labelset[0]
nim = im/np.std(im) # normalize image by std
## test training data ?
inference_train = True
if inference_train:
train_ims, train_segs, train_metas = self.training_set
for tim, tz, tmeta in zip(train_ims, train_segs, train_metas):
## retrieve metadata
islices = tmeta.pop('islices',None)
imask = tmeta.pop('imask', None)
iimask = tmeta.pop('iimask',None)
if islices is not None:
tseeds = self.seeds[islices]
tprior = {
'data': np.asarray(self.prior['data'])[:,iimask],
'imask': imask,
'variance': np.asarray(self.prior['variance'])[:,iimask],
'labelset': self.labelset,
}
if 'intensity' in self.prior:
tprior['intensity'] = self.prior['intensity']
else:
tseeds = self.seeds
tprior = self.prior
## prior
tseg = self.labelset[np.argmax(tz, axis=0)].reshape(tim.shape)
tanchor_api = MetaAnchor(
tprior,
self.prior_functions,
weights_priors,
image=tim,
)
tsol,ty = rwsegment.segment(
tim,
tanchor_api,
seeds=tseeds,
weight_function=weight_function,
**self.rwparams_inf
)
## compute Dice coefficient
tdice = compute_dice_coef(tsol, tseg, labelset=self.labelset)
logger.info('Dice coefficients for train: \n{}'.format(tdice))
nlabel = len(self.labelset)
tflatmask = np.zeros(ty.shape, dtype=bool)
tflatmask[:,imask] = True
loss0 = loss_functions.ideal_loss(tz,ty,mask=tflatmask)
logger.info('Tloss = {}'.format(loss0))
## loss2: squared difference with ztilde
loss1 = loss_functions.anchor_loss(tz,ty,mask=tflatmask)
logger.info('SDloss = {}'.format(loss1))
## loss3: laplacian loss
loss2 = loss_functions.laplacian_loss(tz,ty,mask=tflatmask)
logger.info('LAPloss = {}'.format(loss2))
tanchor_api_h = MetaAnchor(
tprior,
self.prior_functions,
weights_priors_h,
image=tim,
)
tsol,ty = rwsegment.segment(
tim,
tanchor_api_h,
seeds=tseeds,
weight_function=weight_function_h,
**self.rwparams_inf
)
## compute Dice coefficient
tdice = compute_dice_coef(tsol, tseg, labelset=self.labelset)
logger.info('Dice coefficients for train (hand-tuned): \n{}'.format(tdice))
loss0 = loss_functions.ideal_loss(tz,ty,mask=tflatmask)
logger.info('Tloss (hand-tuned) = {}'.format(loss0))
## loss2: squared difference with ztilde
loss1 = loss_functions.anchor_loss(tz,ty,mask=tflatmask)
logger.info('SDloss (hand-tuned) = {}'.format(loss1))
## loss3: laplacian loss
loss2 = loss_functions.laplacian_loss(tz,ty,mask=tflatmask)
logger.info('LAPloss (hand-tuned) = {}'.format(loss2))
break
## prior
anchor_api = MetaAnchor(
self.prior,
self.prior_functions,
weights_priors,
image=nim,
)
sol,y = rwsegment.segment(
nim,
anchor_api,
seeds=self.seeds,
weight_function=weight_function,
**self.rwparams_inf
)
## compute Dice coefficient
dice = compute_dice_coef(sol, seg,labelset=self.labelset)
logger.info('Dice coefficients: \n{}'.format(dice))
## objective
en_rw = rwsegment.energy_rw(
nim, y, seeds=self.seeds,weight_function=weight_function, **self.rwparams_inf)
en_anchor = rwsegment.energy_anchor(
nim, y, anchor_api, seeds=self.seeds, **self.rwparams_inf)
obj = en_rw + en_anchor
logger.info('Objective = {:.3}'.format(obj))
## compute losses
z = seg.ravel()==np.c_[self.labelset]
mask = self.seeds < 0
flatmask = mask.ravel()*np.ones((len(self.labelset),1))
## loss 0 : 1 - Dice(y,z)
loss0 = loss_functions.ideal_loss(z,y,mask=flatmask)
logger.info('Tloss = {}'.format(loss0))
## loss2: squared difference with ztilde
loss1 = loss_functions.anchor_loss(z,y,mask=flatmask)
logger.info('SDloss = {}'.format(loss1))
## loss3: laplacian loss
loss2 = loss_functions.laplacian_loss(z,y,mask=flatmask)
logger.info('LAPloss = {}'.format(loss2))
## loss4: linear loss
loss3 = loss_functions.linear_loss(z,y,mask=flatmask)
logger.info('LINloss = {}'.format(loss3))
## saving
if self.debug:
pass
elif self.isroot:
outdir = self.dir_inf + test_dir
logger.info('saving data in: {}'.format(outdir))
if not os.path.isdir(outdir):
os.makedirs(outdir)
#io_analyze.save(outdir + 'im.hdr',im.astype(np.int32))
#np.save(outdir + 'y.npy',y)
#io_analyze.save(outdir + 'sol.hdr',sol.astype(np.int32))
np.savetxt(outdir + 'objective.txt', [obj])
np.savetxt(
outdir + 'dice.txt',
np.c_[dice.keys(),dice.values()],fmt='%d %f')
f = open(outdir + 'losses.txt', 'w')
f.write('ideal_loss\t{}\n'.format(loss0))
f.write('anchor_loss\t{}\n'.format(loss1))
f.write('laplacian_loss\t{}\n'.format(loss2))
f.close()
def load_or_compute_classifier(self, train, test, mask=None):
from rwsegment import boundary_utils
reload(boundary_utils)
#idir_prior = config.dir_prior_edges + test
#if not os.path.isdir(dir_prior):
# os.makedirs(dir_prior)
## Train classifier
logger.info('train classifier with train {} for test {}'.format(
train, test))
##load image and seg
im = io_analyze.load(config.dir_reg + test + train +
'reggray.hdr').astype(float)
nim = im / np.std(im)
seg = io_analyze.load(config.dir_reg + test + train + 'regseg.hdr')
seg.flat[~np.in1d(seg, self.labelset)] = self.labelset[0]
## sample points
points = boundary_utils.sample_points(im, self.step, mask=mask)
logger.debug('number of sampled points = {}'.format(len(points)))
#impoints = np.zeros(im.shape,dtype=int)
#impoints[tuple(points.T)] = np.arange(len(points)) + 1
## compute edges
edges, edgev, labels = boundary_utils.get_edges(im, points, mask=mask)
logger.debug('number of edges = {}'.format(len(edges)))
## extract profiles
profiles, emap, dists = boundary_utils.get_profiles(nim,
points,
edges,
rad=0)
logger.debug('extracted profiles')
## make features
x = boundary_utils.make_features(
profiles,
size=self.sizex,
additional=[dists, edgev, edgev / dists])
logger.debug('features made, size = {}'.format(len(x[0])))
## make annotations
z = boundary_utils.is_boundary(points, edges, seg)
logger.debug('annotations made')
## learn profiles
logger.debug('training classifier')
classifier = boundary_utils.Classifier()
classifier.train(x, z)
## test classification
logger.debug('testing classifier')
cl, scores = classifier.classify(x)
logger.info('non boundary correct rate: {:.3}'.format(
np.sum((np.r_[cl] == 0) & (np.r_[z] == 0)) /
np.sum(np.r_[z] == 0).astype(float)))
logger.info('boundary correct rate: {:.3}'.format(
np.sum((np.r_[cl] == 1) & (np.r_[z] == 1)) /
np.sum(np.r_[z] == 1).astype(float)))
## store classifier
#np.savetxt(dir_prior + 'classifier.txt', classifier.w)
return classifier
