def __init__(self, model, collectioninst, **kw):
self.resultsfile = argget(kw, 'resultsfile', 'temp.csv')
self.evaluate_uncertainty_times = argget(kw, "evaluate_uncertainty_times", 1)
self.evaluate_uncertainty_dropout = argget(kw, "evaluate_uncertainty_dropout",
1.0) # these standard values ensure that we dont evaluate uncertainty if nothing was provided.
super(LargeVolumeLocationCoordinateClassificationEvaluation, self).__init__(model, collectioninst,
**kw)
def __init__(self, collectioninst, **kw):
self.origargs = copy.deepcopy(kw)
self.use_tensorboard = argget(kw, "use_tensorboard", True, keep=True)
if self.use_tensorboard:
self.image_summaries_each = argget(kw, 'image_summaries_each', 100)
self.dropout_rate = argget(kw, "dropout_rate", 0.5)
self.current_epoch = 0
self.current_iteration = 0
self.restore_optimistically = argget(kw, 'restore_optimistically',
False)
def __init__(self, *w, **kw):
super(CGRUCell, self).__init__(*w, **kw)
self.bnx = argget(kw, "add_x_bn", False)
self.bnh = argget(kw, "add_h_bn", False)
self.bna = argget(kw, "add_a_bn", False)
self.m = argget(kw, "m", None)
self.istraining = argget(kw, 'istraining', tf.constant(True))
self.resgrux = argget(kw, "resgrux", False)
self.resgruh = argget(kw, "resgruh", False)
self.put_r_back = argget(kw, "put_r_back", False)
self.regularize_state = argget(kw, 'use_dropconnect_on_state', False)
filter_shape_h_candidate = self.filter_size_h + [self._num_units] * 2
filter_shape_h_gates = deepcopy(filter_shape_h_candidate)
filter_shape_h_gates[-1] *= 2
filter_shape_x_candidate = self.filter_size_x + [
self.myshapes[0][-1], self._num_units
]
filter_shape_x_gates = deepcopy(filter_shape_x_candidate)
filter_shape_x_gates[-1] *= 2
self.dc_h_factor_gates = self._get_dropconnect(filter_shape_h_gates,
self.dropconnecth,
"mydropconnecthgates")
self.dc_h_factor_candidate = self._get_dropconnect(
filter_shape_h_candidate, self.dropconnecth,
"mydropconnecthcandidate")
self.dc_x_factor_gates = self._get_dropconnect(filter_shape_x_gates,
self.dropconnectx,
"mydropconnectxgates")
self.dc_x_factor_candidate = self._get_dropconnect(
filter_shape_x_candidate, self.dropconnectx,
"mydropconnectxcandidate")
def __init__(self, myshape, num_units, activation=tf.nn.tanh, **kw):
super(CRNNCell, self).__init__()
self._activation = activation
self._num_units = num_units
self.gate = argget(kw, 'gate', sigmoid)
self.periodicconvolution_x = argget(kw, 'periodicconvolution_x', False)
self.periodicconvolution_h = argget(kw, 'periodicconvolution_h', False)
self.filter_size_x = argget(kw, 'filter_size_x', [7, 7])
self.filter_size_h = argget(kw, 'filter_size_h', [7, 7])
self.use_bernoulli = argget(kw, 'use_bernoulli', False)
self.dropconnectx = argget(kw, "dropconnectx", None)
self.dropconnecth = argget(kw, "dropconnecth", None)
self.strides = argget(kw, "strides", None)
if myshape is None:
raise Exception('myshape cant be None!')
myshapein = deepcopy(myshape)
myshapein.pop(-2)
myshapeout = deepcopy(myshape)
myshapeout.pop(-2)
myshapeout[-1] = self._num_units
if self.strides is not None:
if len(myshapeout[1:-1]) != len(self.strides):
raise Exception(
'stride shape should match myshapeout[1:-1]! strides: {}, myshape: {}'.format(self.strides,
myshapeout))
myshapeout[1:-1] = [int(np.round((myshapeout[1 + si]) / self.strides[si])) for si in
range(len(myshapeout) - 2)]
self.myshapes = (myshapein, myshapeout)
def __init__(self, model, collectioninst, **kw):
self.evaluate_uncertainty_times = argget(kw,
"evaluate_uncertainty_times",
1)
self.evaluate_uncertainty_dropout = argget(
kw, "evaluate_uncertainty_dropout", 1.0
) # these standard values ensure that we dont evaluate uncertainty if nothing was provided.
self.evaluate_uncertainty_saveall = argget(
kw, "evaluate_uncertainty_saveall", False)
super(LargeVolumeEvaluation, self).__init__(model, collectioninst,
**kw)
def __init__(self,
featurefiles,
maskfiles=[],
location=None,
tps=None,
**kw):
super(ThreadedGridDataCollection,
self).__init__(featurefiles, maskfiles, location, tps, **kw)
self._batchsize = argget(kw, 'batchsize', 1)
self.num_threads = argget(kw, 'num_threads', 1)
self.curr_thread = 0
self._batch = [None for _ in range(self.num_threads)]
self._batchlabs = [None for _ in range(self.num_threads)]
self._preloadthreads = [
Thread(target=self._preload_random_sample,
args=(
self._batchsize,
it,
)) for it in range(self.num_threads)
]
for t in self._preloadthreads:
t.start()
def create_example_nifti_data_2d(**kw):
shape = argget(kw, "shape", (256, 256))
border_edges = argget(kw, "border_edges", [[64, 192], [64, 192]])
edge_variation = argget(kw, "edge_variation", (25, 25))
rater_variation = argget(kw, "rater_variation", (5, 5))
patients = argget(kw, "patients",
["ruedi", "hans", "eva", "micha", "joerg", "maya", "frieda", "anna", "chelsea", "flynn"])
patient_belongs_to = argget(kw, "patient_belongs_to",
["train", "train", "train", "train", "train", "val", "val", "test", "test", "test"])
testdatadir = argget(kw, "testdatadir", ".")
affine = np.eye(4)
print("creating new testdata at ", testdatadir)
datafolder = argget(kw, "datafolder", "nifti")
testdatadir = os.path.join(testdatadir, datafolder)
if os.path.exists(testdatadir):
print("Files have already been generated. If something is amiss, delete the nifti folder and start again!")
return
for f, pat in zip(patient_belongs_to, patients):
patdir = os.path.join(os.path.join(testdatadir, f), pat)
if not os.path.exists(patdir):
os.makedirs(patdir)
gt_mask = np.zeros(shape)
gt_borders = [[x[0] + (np.random.random() * 2 - 1) * e,
x[1] + (np.random.random() * 2 - 1) * e] for x, e in zip(border_edges, edge_variation)]
gt_borders = np.uint32(gt_borders)
gt_mask[
gt_borders[0][0]:gt_borders[0][1],
gt_borders[1][0]:gt_borders[1][1],
] = 1
for file in ["flair", "mprage", "t2", "pd"]:
myfile = os.path.join(patdir, file + ".nii.gz")
if not os.path.exists(os.path.join(patdir, file + ".nii.gz")):
dat = np.float32(np.random.random(shape)) * np.random.randint(200, 2400) + np.random.randint(200, 800)
dat += gt_mask * np.random.random(shape) * np.random.randint(200, 400)
nib.save(nib.Nifti1Image(dat, affine), myfile)
for file in ["mask1", "mask2"]:
myfile = os.path.join(patdir, file + ".nii.gz")
if not os.path.exists(os.path.join(patdir, file + ".nii.gz")):
dat = np.zeros(shape, dtype=np.uint8)
rater_borders = [[x[0] + (np.random.random() * 2 - 1) * e,
x[1] + (np.random.random() * 2 - 1) * e] for x, e in zip(gt_borders, rater_variation)]
rater_borders = np.uint32(rater_borders)
dat[rater_borders[0][0]: rater_borders[0][1],
rater_borders[1][0]: rater_borders[1][1]] = 1
nib.save(nib.Nifti1Image(dat, affine), myfile)
def __init__(self,
featurefiles,
maskfiles=[],
location=None,
tps=None,
**kw):
super(GridDataCollection, self).__init__(**kw)
self.origargs.update({
"featurefiles": featurefiles,
"maskfiles": maskfiles,
"location": location,
"tps": tps
})
if not isinstance(featurefiles, list):
featurefiles = [featurefiles]
if not isinstance(maskfiles, list):
maskfiles = [maskfiles]
if tps is not None:
self.tps = tps
elif location is not None:
if callable(location):
self.tps = [location]
else:
self.tps = DataCollection.get_all_tps(location, featurefiles,
maskfiles)
if len(self.tps) == 0:
raise Exception(
'no timepoints at location {} containing both {} and {}'.
format(location, maskfiles, featurefiles))
else:
raise Exception('either tps or location has to be set')
if len(self.tps) == 0:
raise Exception(
'there were no timepoints provided and location was not set')
self.featurefiles = featurefiles
self.maskfiles = maskfiles
w = argget(kw, 'w', self.w)
if not isinstance(w, list):
w = np.ndarray.tolist(w)
self.w = w
self.p = argget(kw, 'padding', np.zeros(np.shape(w)))
self.deform = argget(kw, 'deformation', np.zeros(np.shape(w)))
self.interpolate_always = argget(kw, 'interpolate_always', False)
self.deformrandomstate = np.random.RandomState(
argget(kw, 'deformseed', 1234))
self.deformSigma = argget(kw, 'deformSigma', 5)
if not np.isscalar(
self.deformSigma) and len(self.deformSigma) != len(w):
raise Exception(
'we need the same sized deformsigma as w (hence, if we provide an array, it has to be the exact correct size)'
)
self.deformpadding = 2
self.datainterpolation = argget(kw, 'datainterpolation', 3)
self.dataextrapolation = argget(kw, 'dataextrapolation', 'constant')
self.scaling = np.float32(argget(kw, 'scaling', np.zeros(np.shape(w))))
self.rotation = np.float32(argget(kw, 'rotation', 0))
self.shift = np.float32(argget(kw, 'shift', np.zeros(np.shape(w))))
self.mirror = np.float32(argget(kw, 'mirror', np.zeros(np.shape(w))))
self.gaussiannoise = np.float32(argget(kw, 'gaussiannoise', 0.0))
self.vary_mean = np.float32(argget(kw, 'vary_mean', 0))
self.vary_stddev = np.float32(argget(kw, 'vary_stddev', 0))
self.regression = argget(kw, 'regression', False)
self.softlabels = argget(kw, 'softlabels', True)
self.whiten = argget(kw, "whiten", True)
self.each_with_labels = argget(kw, "each_with_labels", 0)
if self.each_with_labels > 0 and len(self.maskfiles) == 0:
raise Exception(
'need to provide at leas tone mask file, otherwise we cant make sure we have labels set obviously'
)
self.whiten_subvolumes = argget(kw, "whiten_subvolumes", False)
self.presize_for_normalization = argget(kw,
'presize_for_normalization',
[None for w in self.w])
self.half_gaussian_clip = argget(kw, 'half_gaussian_clip', False)
self.pyramid_sampling = argget(kw, 'pyramid_sampling', False)
self.subtractGauss = argget(kw, "subtractGauss", False)
self.nooriginal = argget(kw, 'nooriginal', False)
self.subtractGaussSigma = np.float32(argget(kw, "sigma", 5))
self.choose_mask_at_random = argget(kw, "choose_mask_at_random", False)
if self.choose_mask_at_random:
self.random_mask_state = np.random.RandomState(
argget(kw, 'randommaskseed', 1337))
self.zero_out_label = argget(kw, 'zero_out_label', None)
self.running_mean = 0
self.running_num = 0
self.running_var = 0
self.lazy = argget(kw, 'lazy', True)
self.imagedict = {}
self.perform_one_hot_encoding = argget(kw, 'perform_one_hot_encoding',
True)
self.correct_nifti_orientation = argget(kw,
'correct_nifti_orientation',
len(self.w) == 3)
if self.correct_nifti_orientation and len(self.w) != 3:
self.correct_nifti_orientation = False
logging.getLogger('data').warning(
'Can only correct for orientation for 3d data so far!')
self.numoffeatures = argget(
kw, 'numoffeatures',
len(self._get_features_and_masks(self.tps[0])[0]))
self.sample_counter = 0
self.minlabel = argget(kw, 'minlabel', 1)
if self.lazy == False and argget(kw, 'preloadall', False):
self.preload_all()
def __init__(self, model, collectioninst, **kw):
self.nclasses = argget(kw, 'ncoords', [64, 64, 64], keep=True)
super(LocationClassificationEvaluation, self).__init__(model, collectioninst,
**kw)
def remove_example_nifti_data_2d(**kw):
testdatadir = argget(kw, "testdatadir", ".")
datafolder = argget(kw, "datafolder", "nifti")
shutil.rmtree(os.path.join(testdatadir, datafolder))
def __init__(self, model, collectioninst, **kw):
self.nclasses = argget(kw, 'nclasses', 2, keep=True)
super(ClassificationEvaluation, self).__init__(model, collectioninst,
**kw)
def __init__(self, **kw):
self.origargs = copy.deepcopy(kw)
self.randomstate = np.random.RandomState(argget(kw, "seed", 12345678))
self.nclasses = argget(kw, 'nclasses', 2)
def mdgru_bb(self,
inp,
dropout,
num_hidden,
num_output,
noactivation=False,
name=None,
**kw):
dimensions = argget(kw, "dimensions", None)
if dimensions is None:
dimensions = [
i + 1 for i, v in enumerate(inp.get_shape()[1:-1]) if v > 1
]
bnx = argget(kw, "bnx", self.bnx)
bnh = argget(kw, "bnh", self.bnh)
bna = argget(kw, "bna", self.bna)
bne = argget(kw, "bne", self.bne)
resmdgru = argget(kw, 'resmdgru', self.resmdgru)
use_dropconnectx = argget(kw, "use_dropconnectx",
self.use_dropconnectx)
use_dropconnecth = argget(kw, "use_dropconnecth",
self.use_dropconnecth)
cgru_activation = argget(kw, 'cgru_activation', self.cgru_activation)
myMDGRU = MDGRU
with tf.variable_scope(name):
mdgruclass = myMDGRU(
inp,
dropout,
dimensions,
num_hidden=num_hidden,
name="mdgru",
bnx=bnx,
bnh=bnh,
bna=bna,
use_dropconnectx=use_dropconnectx,
use_dropconnecth=use_dropconnecth,
resgrux=self.resgrux,
resgruh=self.resgruh,
m=self.m,
return_cgru_results=self.return_cgru_results,
swap_memory=self.swap_memory,
put_r_back=self.put_r_back,
cgru_activation=cgru_activation,
use_static_rnn=self.use_static_rnn,
no_avgpool=self.no_avgpool,
filter_size_x=self.filter_size_x,
filter_size_h=self.filter_size_h,
use_dropconnect_on_state=self.use_dropconnect_on_state,
legacy_cgru_addition=self.legacy_cgru_addition,
**kw)
mdgru = mdgruclass()
if num_output is not None:
mdgruinnershape = mdgru.get_shape()[1:-1].as_list()
doreshape = False
if len(mdgruinnershape) >= 3:
newshape = [
-1,
np.prod(mdgruinnershape),
mdgru.get_shape().as_list()[-1]
]
mdgru = tf.reshape(mdgru, newshape)
doreshape = True
num_input = mdgru.get_shape().as_list()[-1]
filtershape = [1 for _ in mdgru.get_shape()[1:-1]
] + [num_input, num_output]
numelem = (num_output + num_input) / 2
uniform = False
if self.activation in [tf.nn.elu, tf.nn.relu]:
numelem = (num_input) / 2
uniform = False
W = tf.get_variable("W",
filtershape,
dtype=tf.float32,
initializer=get_modified_xavier_method(
numelem, uniform))
b = tf.get_variable('b', [num_output],
initializer=tf.constant_initializer(0))
mdgru = tf.nn.convolution(mdgru, W, padding="SAME")
if resmdgru:
if doreshape:
inp = tf.reshape(inp, [
-1,
np.prod(inp.get_shape()[1:-1].as_list()),
inp.get_shape().as_list()[-1]
])
resW = tf.get_variable(
'resW', [1 for _ in inp.get_shape().as_list()[1:-1]] +
[inp.get_shape().as_list()[-1], num_output],
dtype=tf.float32,
initializer=get_modified_xavier_method(
num_output, False))
mdgru = tf.nn.convolution(inp, resW,
padding="SAME") + mdgru
if bne:
mdgru = batch_norm(mdgru,
"bne",
mdgruclass.istraining,
bias=False,
m=mdgruclass.m)
mdgru = mdgru + b
if doreshape:
mdgru = tf.reshape(mdgru, [-1] + mdgruinnershape +
[mdgru.get_shape().as_list()[-1]])
if noactivation:
return mdgru
else:
return self.activation(mdgru)
def __init__(self, data, target, dropout, **kw):
super(MDGRUNet, self).__init__()
self.bnx = argget(kw, "bnx", False)
self.bnh = argget(kw, "bnh", False)
self.bna = argget(kw, "bna", False)
self.bne = argget(kw, "bne", False)
self.use_dropconnectx = argget(kw, "use_dropconnectx", True)
self.use_dropconnecth = argget(kw, "use_dropconnecth", False)
self.resmdgru = argget(kw, "resmdgru", False)
self.resgrux = argget(kw, 'resgrux', False)
self.resgruh = argget(kw, 'resgruh', False)
self.m = argget(kw, "m", None)
self.swap_memory = argget(kw, "swap_memory", False)
self.return_cgru_results = argget(kw, "return_cgru_results", False)
self.put_r_back = argget(kw, "put_r_back", False)
self.use_static_rnn = argget(kw, 'use_static_rnn', False)
self.no_avgpool = argget(kw, 'no_avgpool', True)
self.filter_size_x = argget(kw, 'filter_size_x', [7, 7, 7])
self.filter_size_h = argget(kw, 'filter_size_h', [7, 7, 7])
self.cgru_activation = argget(kw, 'rnn_activation', tf.nn.tanh)
self.activation = argget(kw, 'activation', tf.nn.tanh)
self.use_caffe_impl = argget(kw, "use_caffe_impl", False)
self.favor_speed_over_memory = argget(kw, "favor_speed_over_memory",
True)
self.use_dropconnect_on_state = argget(kw, 'use_dropconnect_on_state',
False)
self.legacy_cgru_addition = argget(kw, 'legacy_cgru_addition', False)
def __init__(self, evaluationinstance, **kw):
self.origargs = copy.deepcopy(kw)
# prelogging:
experiments = argget(kw, 'experimentloc',
os.path.expanduser('~/experiments'))
self.runfile = [
f[1] for f in inspect.stack() if re.search("RUN.*\.py", f[1])
][0]
self.experiments_postfix = argget(kw, 'experiments_postfix', "")
experiments_nots = os.path.join(
experiments,
'{}'.format(self.runfile[self.runfile.index("RUN_") + 4:-3] +
self.experiments_postfix))
self.experiments = os.path.join(experiments_nots,
str(int(time.time())))
self.cachefolder = os.path.join(self.experiments, 'cache')
os.makedirs(self.cachefolder)
# logging:
loggers = [
logging.getLogger(n)
for n in ['model', 'eval', 'runner', 'helper', 'data']
]
formatter = logging.Formatter(
'%(asctime)s %(name)s\t%(levelname)s:\t%(message)s')
logfile = argget(kw, 'logfile',
os.path.join(self.cachefolder, 'log.txt'))
fh = logging.FileHandler(logfile)
fh.setLevel(argget(kw, 'logfileloglvl', logging.DEBUG))
fh.setFormatter(formatter)
self.only_cpu = argget(kw, 'only_cpu', False)
ch = logging.StreamHandler()
ch.setFormatter(formatter)
ch.setLevel(argget(kw, 'loglvl', logging.WARNING))
for logger in loggers:
logger.addHandler(ch)
logger.setLevel(logging.DEBUG)
logger.addHandler(fh)
for k in self.origargs:
logging.getLogger('runner').info('arg {}:{}'.format(
k, self.origargs[k]))
self.ev = evaluationinstance
for k in self.ev.origargs:
logging.getLogger('eval').info('arg {}:{}'.format(
k, self.ev.origargs[k]))
for k in self.ev.trdc.origargs:
logging.getLogger('data').info(' trdc arg {}:{}'.format(
k, self.ev.trdc.origargs[k]))
for k in self.ev.tedc.origargs:
logging.getLogger('data').info(' tedc arg {}:{}'.format(
k, self.ev.tedc.origargs[k]))
for k in self.ev.valdc.origargs:
logging.getLogger('data').info('valdc arg {}:{}'.format(
k, self.ev.valdc.origargs[k]))
for k in self.ev.model.origargs:
logging.getLogger('model').info('arg {}:{}'.format(
k, self.ev.model.origargs[k]))
self.episodes = argget(kw, 'episodes', ['train', 'evaluate'])
self.epochs = argget(kw, 'epochs', 1)
self.batch_size = argget(kw, 'batch_size', 1)
self.its_per_epoch = argget(
kw, 'its_per_epoch',
self.ev.trdc.get_data_dims()[0] // self.batch_size)
self.checkpointfiles = argget(kw, 'checkpointfiles', None)
self.estimatefilenames = argget(kw, 'estimatefilenames', None)
if isinstance(self.checkpointfiles, list):
if 'train' in self.episodes:
logging.getLogger('runner').error(
'Multiple checkpoints are only allowed if only testing is performed.'
)
exit(1)
else:
self.checkpointfiles = [self.checkpointfiles]
if not isinstance(self.estimatefilenames, list):
self.estimatefilenames = [self.estimatefilenames]
if len(self.checkpointfiles) != len(self.estimatefilenames):
if len(self.estimatefilenames) != 1:
logging.getLogger('runner').error(
'Optionnames must match number of checkpoint files or have length 1!'
)
exit(1)
else:
self.estimatefilenames = [
self.estimatefilenames[0] +
"-{}-{}".format(i, os.path.basename(c))
for i, c in enumerate(self.checkpointfiles)
]
self.plotfolder = os.path.join(self.experiments, 'plot')
self.plot_scaling = argget(kw, 'plot_scaling', 1e-8)
self.display_each = argget(kw, 'display_each', 100)
self.test_each = argget(kw, 'test_each', self.display_each)
self.save_each = argget(kw, 'save_each', self.display_each)
self.plot_each = argget(kw, 'plot_each', self.display_each)
self.test_size = argget(kw, 'test_size', 1) # batch_size for tests
self.test_iters = argget(kw, 'test_iters', 1)
self._test_pick_iteration = argget(kw,
'test_first',
ifset=0,
default=self.test_each - 1)
self.perform_n_times_full_validation = argget(
kw, 'perform_n_times_full_validation', 0)
self.perform_n_times_full_validation_dropout = argget(
kw, 'perform_n_times_full_validation_dropout', 0.5)
self.show_testing_results = argget(kw, 'show_testing_results', False)
force_symlink(self.experiments, os.path.join(experiments_nots,
"latest"))
os.makedirs(self.plotfolder)
self.printIt = argget(kw, "print_testing_results", True)
self.gpubound = argget(kw, 'gpubound', 1)
self.notifyme = argget(kw, 'notifyme', None)
self.results_to_csv = argget(kw, 'results_to_csv', False)
self.train_losses = []
self.test_losses = []
self.val_losses = []
# remove full parameters since it would not be an ignored parameter and confuse us
kw.pop('fullparameters')
if kw:
logging.getLogger('runner').warning(
'the following args were ignored: {}'.format(kw))
def __init__(self,
inputarr,
dropout,
dimensions=None,
layers=1,
num_hidden=100,
name="mdgru",
**kw):
'''
@param inputarr: needs to be in batch, spatialdim1...spatialdimn,channel form
'''
self.inputarr = inputarr
self.add_x_bn = argget(kw, "bnx", False)
self.add_h_bn = argget(kw, "bnh", False)
self.add_a_bn = argget(kw, "bna", False)
self.istraining = argget(kw, 'istraining', tf.constant(True))
if dimensions is None:
self.dimensions = [
x + 1 for x in range(len(inputarr.get_shape()[1:-1]))
]
else:
self.dimensions = dimensions
self.layers = layers
self.num_hidden = num_hidden
self.name = name
self.dropout = dropout
self.use_dropconnectx = argget(kw, "use_dropconnectx", True)
self.use_dropconnecth = argget(kw, "use_dropconnecth", False)
self.use_bernoulli = argget(kw, 'use_bernoulli_dropconnect', False)
self.m = argget(kw, "min_mini_batch", None)
self.resgruh = argget(kw, "resgruh", False)
self.resgrux = argget(kw, "resgrux", False)
self.filter_size_x = argget(kw, 'filter_size_x', [7, 7, 7])
self.filter_size_h = argget(kw, 'filter_size_h', [7, 7, 7])
self.return_cgru_results = argget(kw, 'return_cgru_results', False)
self.use_dropconnect_on_state = argget(kw, 'use_dropconnect_on_state',
False)
self.strides = argget(kw, "strides", None)
self.swap_memory = argget(kw, "swap_memory", True)
self.put_r_back = argget(kw, "put_r_back", False)
self.cgru_activation = argget(kw, 'cgru_activation', tf.nn.tanh)
self.use_static_rnn = argget(kw, 'use_static_rnn', False)
self.no_avgpool = argget(kw, 'no_avgpool', True)
self.legacy_cgru_addition = argget(kw, 'legacy_cgru_addition', False)
def __init__(self,
inputarr,
dropout,
dimensions=None,
layers=1,
num_hidden=100,
name="mdgru",
**kw):
self.inputarr = inputarr
self.generator = argget(kw, "generator", False)
self.add_x_bn = argget(kw, "bnx", False)
self.add_h_bn = argget(kw, "bnh", False)
self.add_a_bn = argget(kw, "bna", False)
self.form = argget(kw, "form", "NDHWC")
self.istraining = argget(kw, 'istraining', tf.constant(True))
if dimensions is None:
self.dimensions = [
x + 1 for x in range(len(inputarr.get_shape()[1:-1]))
]
else:
self.dimensions = dimensions
self.layers = layers
self.num_hidden = num_hidden
self.name = name
self.dropout = dropout
self.use_dropconnectx = argget(kw, "use_dropconnectx", True)
self.use_dropconnecth = argget(kw, "use_dropconnecth", False)
self.use_bernoulli = argget(kw, 'use_bernoulli_dropconnect', False)
self.mask_padding = argget(kw, 'maskpadding', None)
self.m = argget(kw, "min_mini_batch", None)
self.favor_speed_over_memory = argget(kw, "favor_speed_over_memory",
False)
self.filter_sizes = argget(kw, 'filter_sizes', [7, 7, 7])
if any([self.add_x_bn, self.add_h_bn, self.add_a_bn]):
raise Exception("bn not allowed for caffemdgru")
def __init__(self, model, collectioninst, **kw):
super(SupervisedEvaluation, self).__init__(collectioninst, **kw)
self.setupCollections(collectioninst)
self.currit = 0
self.namespace = argget(kw, "namespace", "default")
self.only_save_labels = argget(kw, "only_save_labels", False)
with tf.variable_scope(self.namespace):
self.training = tf.placeholder(dtype=tf.bool)
self.dropout = tf.placeholder(dtype=tf.float32)
self.data = tf.placeholder(dtype=tf.float32,
shape=self.trdc.get_shape())
if type(self.nclasses) == list: # for location classification
self.target = tf.placeholder(
dtype=tf.float32,
shape=self.trdc.get_target_shape()[:-1] +
[np.sum(self.nclasses)])
else:
self.target = tf.placeholder(
dtype=tf.float32,
shape=self.trdc.get_target_shape()[:-1] + [self.nclasses])
kw_copy = copy.deepcopy(kw)
self.model = model(self.data,
self.target,
self.dropout,
training=self.training,
**kw)
self.model.optimize
# in the case we have a different testing set, we can construct 2 graphs, one for training and testing case
if self.tedc.get_shape() != self.trdc.get_shape():
self.test_graph = tf.Graph()
with self.test_graph.as_default():
with tf.variable_scope(self.namespace):
self.test_training = tf.placeholder(dtype=tf.bool)
self.test_dropout = tf.placeholder(dtype=tf.float32)
self.test_data = tf.placeholder(
dtype=tf.float32, shape=self.tedc.get_shape())
if type(self.nclasses
) == list: # for location classification
self.test_target = tf.placeholder(
dtype=tf.float32,
shape=self.tedc.get_target_shape()[:-1] +
[np.sum(self.nclasses)])
else:
self.test_target = tf.placeholder(
dtype=tf.float32,
shape=self.tedc.get_target_shape()[:-1] +
[self.nclasses])
self.test_model = model(self.test_data,
self.test_target,
self.test_dropout,
training=self.test_training,
**kw_copy)
self.test_model.prediction
self.test_model.cost
else:
self.test_graph = tf.get_default_graph()
self.test_model = self.model
self.test_training = self.training
self.test_dropout = self.dropout
self.test_data = self.data
self.test_target = self.target
self.batch_size = argget(kw, 'batch_size', 1)
self.validate_same = argget(kw, 'validate_same', False)
