def montecarlo(cause, effect, unknown, n, *ignore):
cnt_cause = count(zip(cause))
cnt_unknown = count(zip(unknown))
cnt_cause_effect = count(zip(cause, effect))
cnt_effect_unknown = count(zip(effect, unknown))
sumarr(cnt_cause_effect, 0.1) # make beta dist work with zeros
sumarr(cnt_cause, 0.1)
sumarr(cnt_unknown, 0.1)
sumarr(cnt_effect_unknown, 0.1)
cnt_cause_unknown = count(zip(cause, unknown))
rounds = 500
p_overall = struct(cause_unknown_chain=[[0,0],[0,0]],
cause_unknown_collide=[[0,0],[0,0]])
for i in range(rounds):
p=struct()
p.cause = 1-beta(*cnt_cause)
p.unknown = 1-beta(*cnt_unknown)
p.effect_given_cause = [1-beta(*cnts) for cnts in cnt_cause_effect]
p.unknown_given_effect = [1-beta(*cnts) for cnts in cnt_effect_unknown]
p = get_joints_by_model(p)
acclarr(p_overall.cause_unknown_chain, p.cause_unknown_chain)
acclarr(p_overall.cause_unknown_collide, p.cause_unknown_collide)
mularr(p_overall.cause_unknown_chain, 1.0/rounds)
mularr(p_overall.cause_unknown_collide, 1.0/rounds)
try:
bayes_factor = get_factor(p_overall, cnt_cause_unknown)
except ValueError:
print '==ValueError=='
print p_overall.__dict__
raise ValueError()
return struct(bayes_fwd_rev=bayes_factor)
def mle(cause, effect, unknown, n, p_cause, p_effect_given_cause):
p=struct(cause=p_cause, effect_given_cause=p_effect_given_cause)
cnt = count(zip(effect, unknown))
chi_indep = chi2_contingency(cnt)
p.unknown_given_effect = [ float(cnt[0][1]) / sum(cnt[0]),
float(cnt[1][1]) / sum(cnt[1]) ]
cnt = count(zip(unknown))
p.unknown = float(cnt[1]) / sum(cnt)
p = get_joints_by_model(p)
cnt = count(zip(cause, unknown))
bayes_factor = get_factor(p, cnt)
return struct(reject_indep=chi_indep,
bayes_fwd_rev=bayes_factor)
def chi2_dir(cause, effect, unknown, n, p_cause, p_effect_given_cause):
cnt = count(zip(effect, unknown))
#print cnt
chi_indep = chi2_contingency(cnt)[1]
p_unknown_given_effect = [ float(cnt[0][1]) / sum(cnt[0]),
float(cnt[1][1]) / sum(cnt[1]) ]
#print 'p(bact|cd)=%s' % p_unknown_given_effect
exp=[[0,0],[0,0]]
for c in range(2):
for e in range(2):
for u in range(2):
exp[c][u] += (n *
p_of_val(p_cause, c) *
p_of_val(p_effect_given_cause[c], e) *
p_of_val(p_unknown_given_effect[e], u))
cnt = count(zip(cause, unknown))
#print "obs=%s" % cnt
#print 'cnt=%s' % cnt
#print 'expected if cd->bact=%s' % exp
chi_rev = chisquare(cnt, exp, axis=None, ddof=2)
chi_fwd = chi2_contingency(cnt)
#print 'expected if bact->cd=%s' % chi_fwd[3]
bayes_factor = chi2.pdf(chi_fwd[0],1) / chi2.pdf(chi_rev.statistic,1)
return struct(reject_indep=chi_indep,
bayes_fwd_rev=bayes_factor,
reject_fwd=chi_fwd[1],
reject_rev=chi_rev.pvalue)
def get_caffe_params(l_name):
layer_params = np.array(net.params[l_name])
filter = caffe.io.blobproto_to_array(layer_params[0])
bias = caffe.io.blobproto_to_array(layer_params[1])
return utils.struct(
filter=filter,
bias=bias
)
def mnistloader(mnist_path="../MNIST_data"):
'''
Args :
mnist_path - string
path of mnist folder
'''
mnist = input_data.read_data_sets(mnist_path, one_hot=True)
train = struct()
test = struct()
val = struct()
train.image = mnist.train.images
train.label = mnist.train.labels
test.image = mnist.test.images
test.label = mnist.test.labels
val.image = mnist.validation.images
val.label = mnist.validation.labels
return train, test, val
def get_caffe_params(l_name):
layer_params = np.array(net.params[l_name])
filter = caffe.io.blobproto_to_array(layer_params[0])
bias = caffe.io.blobproto_to_array(layer_params[1])
return utils.struct(
filter=filter,
bias=bias
)
def build_model(batch_size=batch_size):
l_in = nn.layers.InputLayer(shape=(batch_size,)+image.shape)
l = l_in
l = conv3(l, num_filters=64)
l = conv3(l, num_filters=64)
l = max_pool(l)
l = conv3(l, num_filters=128)
l = conv3(l, num_filters=128)
l = max_pool(l)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = max_pool(l)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = max_pool(l)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = max_pool(l)
l = dnn.Conv2DDNNLayer(l,
num_filters=4096,
strides=(1, 1),
border_mode="valid",
filter_size=(7,7))
l = dnn.Conv2DDNNLayer(l,
num_filters=4096,
strides=(1, 1),
border_mode="same",
filter_size=(1,1))
l = dnn.Conv2DDNNLayer(l,
num_filters=n_classes,
strides=(1,1),
border_mode="same",
filter_size=(1,1),
nonlinearity=None)
l_to_strengthen = l
l_out = l
return utils.struct(
input=l_in,
out=l_out,
to_strengthen=l_to_strengthen)
def build_model(batch_size=batch_size):
l_in = nn.layers.InputLayer(shape=(batch_size,)+image.shape)
l = l_in
l = conv3(l, num_filters=64)
l = conv3(l, num_filters=64)
l = max_pool(l)
l = conv3(l, num_filters=128)
l = conv3(l, num_filters=128)
l = max_pool(l)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = max_pool(l)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = max_pool(l)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = max_pool(l)
l = dnn.Conv2DDNNLayer(l,
num_filters=4096,
strides=(1, 1),
border_mode="valid",
filter_size=(7,7))
l = dnn.Conv2DDNNLayer(l,
num_filters=4096,
strides=(1, 1),
border_mode="same",
filter_size=(1,1))
l = dnn.Conv2DDNNLayer(l,
num_filters=n_classes,
strides=(1,1),
border_mode="same",
filter_size=(1,1),
nonlinearity=None)
l_to_strengthen = l
l_out = l
return utils.struct(
input=l_in,
out=l_out,
to_strengthen=l_to_strengthen)
def vgg16(batch_shape):
"""
Create a vgg16, with the parameters from http://www.robots.ox.ac.uk/~vgg/research/very_deep/
See googlenet.py for the method used to convert these caffe parameters to lasagne parameters.
:param batch_shape: The shape of the input images. This should be of size (N, 3, X>=224, Y>=224). Note flexible
image size, as the last dense layers have been implemented here with convolutional layers.
:return: a struct with the input layer, the logit layer (before the final softmax) and the output layer.
"""
l_in = nn.layers.InputLayer(shape=batch_shape)
l = l_in
l = conv3(l, num_filters=64)
l = conv3(l, num_filters=64)
l = max_pool(l)
l = conv3(l, num_filters=128)
l = conv3(l, num_filters=128)
l = max_pool(l)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = max_pool(l)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = max_pool(l)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = max_pool(l)
l = dnn.Conv2DDNNLayer(l, num_filters=4096, stride=(1, 1), border_mode="valid", filter_size=(7, 7))
l = dnn.Conv2DDNNLayer(l, num_filters=4096, stride=(1, 1), border_mode="same", filter_size=(1, 1))
l_logit = dnn.Conv2DDNNLayer(
l, num_filters=1000, stride=(1, 1), border_mode="same", filter_size=(1, 1), nonlinearity=None
)
l_logit_flat = nn.layers.FlattenLayer(l_logit)
l_dense = nn.layers.NonlinearityLayer(l_logit_flat, nonlinearity=nn.nonlinearities.softmax)
filename = os.path.join(os.path.dirname(os.path.abspath(__file__)), "vgg16.pkl")
if os.path.exists(filename):
with open(filename, "r") as f:
nn.layers.set_all_param_values(l_dense, pickle.load(f))
return utils.struct(input=l_in, logit=l_logit, out=l_dense)
def read_from_files(self,
sick_fn='../../study2_sick.txt',
gene_fn='../../study2_genome.txt',
sample_fn='../study2/gsi',
sex_fn='../study2/sex',
bact_dir='../study2/'):
for fn in [sick_fn, gene_fn]:
for line in file(fn):
line = line.strip()
if line=='' or line[0]=='#':
continue
words = line.split('\t')
if words[0]=='dbGaP SubjID':
continue
if fn==sick_fn:
self.patients[words[0]] = struct(id=words[0], health=words[3])
else:
self.patients[words[0]].nod2 = int(words[2])
self.patients[words[0]].atg16l1 = int(words[3])
pid_by_sample = {}
for line in file(sample_fn):
line=line.strip()
[sample, pid]=line.split(' ')
if 'samples' not in self.patients[pid]:
self.patients[pid].samples=[]
self.patients[pid].samples.append(sample)
pid_by_sample[sample]=pid
for line in file(sex_fn):
line=line.strip()
[x, sample, sex] = line.split(' ')
patient = self.patients[pid_by_sample[sample]]
if 'sex' in patient and patient.sex!=sex:
print 'WARNING: inconsistent sex for %s at %s' % (patient.id, sample)
if sex!='unknown':
patient.sex = sex
for pid in self.patients:
patient = self.patients[pid]
if 'samples' not in patient:
continue
patient.raw_counts = defaultdict(lambda:0)
patient.total_count = 0
for sid in patient.samples:
for line in file(bact_dir+sid+'.results'):
try:
(n,species)=re.match(" *([0-9]*) (.*)\n",line).groups()
except:
print 'ERROR: '+line
n=int(n)
patient.raw_counts[species] += n
patient.total_count += n
self.bacteria[species] = True
patient.fractions = defaultdict(lambda:0.0)
for species in patient.raw_counts:
if species > 1:
patient.fractions[species] = float(patient.raw_counts[species]) / patient.total_count
def mnistloader():
'''
Return :
struct is defined in utils.py
train - mnist train set with struct
test - mnist test set with struct
val - mnist validation set with struct
'''
mnist = input_data.read_data_sets(MNIST_PATH, one_hot=True)
train = struct()
test = struct()
val = struct()
train.image = mnist.train.images
train.label = mnist.train.labels
test.image = mnist.test.images
test.label = mnist.test.labels
val.image = mnist.validation.images
val.label = mnist.validation.labels
return train, test, val
def loadLabels(self, j):
shapes = []
for d in j['shapes']:
s = Shape(None, None)
s.__dict__.update(d)
s.label = utils.struct(**s.label)
s.points = [QtCore.QPointF(x, y) for x, y in s.points]
if s.shape_type == Mode_box:
box = Box3D(s)
box.bounds = d['box']
box.label = s.label
s.box = box
s.__class__ = ShapeBox
shapes.append(s)
return shapes
def build_model(batch_size=batch_size, seq_length=seq_length):
l_in = nn.layers.InputLayer(shape=(batch_size, seq_length, n_channels)+im_shape)
l = l_in
l = nn.layers.ReshapeLayer(l, (batch_size*seq_length,n_channels)+im_shape)
l = conv_layer(l, num_filters=16, filter_size=(3,3))
l = conv_layer(l, num_filters=16, filter_size=(3,3))
l = max_pooling(l, ds=(2,2))
l = conv_layer(l, num_filters=32, filter_size=(3,3))
l = conv_layer(l, num_filters=32, filter_size=(3,3))
l = max_pooling(l, ds=(2,2))
l = conv_layer(l, num_filters=64, filter_size=(3,3))
l = conv_layer(l, num_filters=64, filter_size=(3,3))
l = max_pooling(l, ds=(2,2))
l = conv_layer(l, num_filters=128, filter_size=(3,3))
l = conv_layer(l, num_filters=128, filter_size=(3,3))
l = max_pooling(l, ds=(2,2))
l = nn.layers.ReshapeLayer(l, (batch_size, seq_length, -1))
l = nn.layers.DropoutLayer(l, p=0.5)
n_states = 2048
l_fwd = rnn_layer(l, n_states)
l_bck = rnn_layer(l, n_states, backwards=True)
l = nn.layers.ElemwiseSumLayer([l_fwd, l_bck])
l = nn.layers.DropoutLayer(l, p=0.5)
l = nn.layers.ReshapeLayer(l, (batch_size*seq_length, n_states))
l_out = nn.layers.DenseLayer(l, num_units=n_classes,
nonlinearity=nn.nonlinearities.softmax)
l_out_reshape = nn.layers.ReshapeLayer(l_out, (batch_size, seq_length, n_classes))
return utils.struct(
input=l_in,
out=l_out,
out_reshape=l_out_reshape)
def get_googlenet(batchsize=10, n_channels=3, img_height=224, img_width=224):
l_in = lasagne.layers.InputLayer(
shape=(batchsize, n_channels, img_height, img_width),
name='input',
)
l_conv1 = dnn.Conv2DDNNLayer(
l_in,
num_filters=64,
filter_size=(7, 7),
pad=3,
stride=(2, 2),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv1/7x7_s2',
)
l_pool1 = flip(dnn.MaxPool2DDNNLayer(
flip(l_conv1),
pool_size=(3, 3),#pool_size
stride=(2, 2),
pad=(1,1),
name='pool1/3x3_s2'
))
lrn = LocalResponseNormalization2DLayer(
l_pool1,
alpha= 0.0001 / 5,
beta= 0.75,
k=1,
n=5,
name='pool1/norm1',
)
l_conv2 = dnn.Conv2DDNNLayer(
lrn,
num_filters=64,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv2/3x3_reduce',
)
l_conv2b = dnn.Conv2DDNNLayer(
l_conv2,
num_filters=192,
filter_size=(3, 3),
pad=1,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv2/3x3',
)
lrn2 = LocalResponseNormalization2DLayer(
l_conv2b,
alpha= 0.0001 / 5,
beta= 0.75,
k=1,
n=5,
name='conv2/norm2',
)
l_pool2 = flip(dnn.MaxPool2DDNNLayer(
flip(lrn2),
pool_size=(3, 3),#pool_size
stride=(2, 2),
pad=(1,1),
name='pool2/3x3_s2'
))
def inception(layer, name, no_1x1=64, no_3x3r=96, no_3x3=128, no_5x5r=16, no_5x5=32, no_pool=32):
l_conv_inc = dnn.Conv2DDNNLayer(
layer,
num_filters=no_1x1,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name+'/1x1',
)
l_conv_inc2 = dnn.Conv2DDNNLayer(
layer,
num_filters=no_3x3r,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name+'/3x3_reduce',
)
l_conv_inc2b = dnn.Conv2DDNNLayer(
l_conv_inc2,
num_filters=no_3x3,
filter_size=(3, 3),
pad=1,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name+'/3x3',
)
l_conv_inc2c = dnn.Conv2DDNNLayer(
layer,
num_filters=no_5x5r,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name+'/5x5_reduce',
)
l_conv_inc2d = dnn.Conv2DDNNLayer(
l_conv_inc2c,
num_filters=no_5x5,
filter_size=(5, 5),
pad=2,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name+'/5x5',
)
l_pool2 = flip(dnn.MaxPool2DDNNLayer(
flip(layer),
pool_size=(3, 3),#pool_size
stride=(1, 1),
pad=(1, 1),
name=name+'/pool'
))
l_conv_inc2e = dnn.Conv2DDNNLayer(
l_pool2,
num_filters=no_pool,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name+'/pool_proj',
)
l_inc_out = nn.layers.concat([l_conv_inc, l_conv_inc2b, l_conv_inc2d, l_conv_inc2e])
return l_inc_out
l_inc_3a = inception(l_pool2, 'inception_3a', no_1x1=64, no_3x3r=96, no_3x3=128, no_5x5r=16, no_5x5=32, no_pool=32)
l_inc_3b = inception(l_inc_3a, 'inception_3b', no_1x1=128, no_3x3r=128, no_3x3=192, no_5x5r=32, no_5x5=96, no_pool=64)
l_pool3 = flip(dnn.MaxPool2DDNNLayer(
flip(l_inc_3b),
pool_size=(3, 3),#pool_size
stride=(2, 2),
pad=1,
name='pool3/3x3_s2'
))
l_inc_4a = inception(l_pool3, 'inception_4a', no_1x1=192, no_3x3r=96, no_3x3=208, no_5x5r=16, no_5x5=48, no_pool=64)
l_inc_4b = inception(l_inc_4a, 'inception_4b', no_1x1=160, no_3x3r=112, no_3x3=224, no_5x5r=24, no_5x5=64, no_pool=64)
l_inc_4c = inception(l_inc_4b, 'inception_4c', no_1x1=128, no_3x3r=128, no_3x3=256, no_5x5r=24, no_5x5=64, no_pool=64)
l_inc_4d = inception(l_inc_4c, 'inception_4d', no_1x1=112, no_3x3r=144, no_3x3=288, no_5x5r=32, no_5x5=64, no_pool=64)
l_inc_4e = inception(l_inc_4d, 'inception_4e', no_1x1=256, no_3x3r=160, no_3x3=320, no_5x5r=32, no_5x5=128, no_pool=128)
l_pool4 = flip(dnn.MaxPool2DDNNLayer(
flip(l_inc_4e),
pool_size=(3, 3),#pool_size
stride=(2, 2),
pad=1,
name='pool4/3x3_s2'
))
l_inc_5a = inception(l_pool4, 'inception_5a', no_1x1=256, no_3x3r=160, no_3x3=320, no_5x5r=32, no_5x5=128, no_pool=128)
l_inc_5b = inception(l_inc_5a, 'inception_5b', no_1x1=384, no_3x3r=192, no_3x3=384, no_5x5r=48, no_5x5=128, no_pool=128)
l_pool5 = flip(dnn.Pool2DDNNLayer(
flip(l_inc_5b),
pool_size=(7, 7),#pool_size
stride=(1, 1),
pad=0,
mode='average',
name='pool5/7x7_s1'
))
l_logit = dnn.Conv2DDNNLayer(
l_pool5,
num_filters=1000,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.linear,
name='prob',
)
l_logit_flat = lasagne.layers.FlattenLayer(l_logit)
l_dense = lasagne.layers.NonlinearityLayer(l_logit_flat, nonlinearity=lasagne.nonlinearities.softmax)
l_out = l_dense
filename = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'googlenet.pkl')
if os.path.exists(filename):
with open(filename, 'r') as f:
nn.layers.set_all_param_values(l_dense, pickle.load(f))
return utils.struct(
input=l_in,
logit=l_logit,
out=l_out
)
def googlenet(batch_shape):
"""
Create a googlenet, with the parameters from https://github.com/BVLC/caffe/tree/master/models/bvlc_googlenet
:param batch_shape: The shape of the input images. This should be of size (N, 3, 224, 224)
:return: a struct with the input layer, the logit layer (before the final softmax) and the output layer.
"""
l_in = lasagne.layers.InputLayer(
shape=batch_shape,
name='input',
)
l_conv1 = dnn.Conv2DDNNLayer(
l_in,
num_filters=64,
filter_size=(7, 7),
pad=3,
stride=(2, 2),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv1/7x7_s2',
)
l_pool1 = flip(dnn.MaxPool2DDNNLayer(
flip(l_conv1),
pool_size=(3, 3), # pool_size
stride=(2, 2),
pad=(1, 1),
name='pool1/3x3_s2'
))
lrn = LocalResponseNormalization2DLayer(
l_pool1,
alpha=0.0001 / 5,
beta=0.75,
k=1,
n=5,
name='pool1/norm1',
)
l_conv2 = dnn.Conv2DDNNLayer(
lrn,
num_filters=64,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv2/3x3_reduce',
)
l_conv2b = dnn.Conv2DDNNLayer(
l_conv2,
num_filters=192,
filter_size=(3, 3),
pad=1,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv2/3x3',
)
lrn2 = LocalResponseNormalization2DLayer(
l_conv2b,
alpha=0.0001 / 5,
beta=0.75,
k=1,
n=5,
name='conv2/norm2',
)
l_pool2 = flip(dnn.MaxPool2DDNNLayer(
flip(lrn2),
pool_size=(3, 3), # pool_size
stride=(2, 2),
pad=(1, 1),
name='pool2/3x3_s2'
))
def inception(layer, name, no_1x1=64, no_3x3r=96, no_3x3=128, no_5x5r=16, no_5x5=32, no_pool=32):
l_conv_inc = dnn.Conv2DDNNLayer(
layer,
num_filters=no_1x1,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/1x1',
)
l_conv_inc2 = dnn.Conv2DDNNLayer(
layer,
num_filters=no_3x3r,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/3x3_reduce',
)
l_conv_inc2b = dnn.Conv2DDNNLayer(
l_conv_inc2,
num_filters=no_3x3,
filter_size=(3, 3),
pad=1,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/3x3',
)
l_conv_inc2c = dnn.Conv2DDNNLayer(
layer,
num_filters=no_5x5r,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/5x5_reduce',
)
l_conv_inc2d = dnn.Conv2DDNNLayer(
l_conv_inc2c,
num_filters=no_5x5,
filter_size=(5, 5),
pad=2,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/5x5',
)
l_pool2 = flip(dnn.MaxPool2DDNNLayer(
flip(layer),
pool_size=(3, 3), # pool_size
stride=(1, 1),
pad=(1, 1),
name=name + '/pool'
))
l_conv_inc2e = dnn.Conv2DDNNLayer(
l_pool2,
num_filters=no_pool,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/pool_proj',
)
l_inc_out = nn.layers.concat([l_conv_inc, l_conv_inc2b, l_conv_inc2d, l_conv_inc2e])
return l_inc_out
l_inc_3a = inception(l_pool2, 'inception_3a', no_1x1=64, no_3x3r=96, no_3x3=128, no_5x5r=16, no_5x5=32, no_pool=32)
l_inc_3b = inception(l_inc_3a, 'inception_3b', no_1x1=128, no_3x3r=128, no_3x3=192, no_5x5r=32, no_5x5=96,
no_pool=64)
l_pool3 = flip(dnn.MaxPool2DDNNLayer(
flip(l_inc_3b),
pool_size=(3, 3), # pool_size
stride=(2, 2),
pad=1,
name='pool3/3x3_s2'
))
l_inc_4a = inception(l_pool3, 'inception_4a', no_1x1=192, no_3x3r=96, no_3x3=208, no_5x5r=16, no_5x5=48, no_pool=64)
l_inc_4b = inception(l_inc_4a, 'inception_4b', no_1x1=160, no_3x3r=112, no_3x3=224, no_5x5r=24, no_5x5=64,
no_pool=64)
l_inc_4c = inception(l_inc_4b, 'inception_4c', no_1x1=128, no_3x3r=128, no_3x3=256, no_5x5r=24, no_5x5=64,
no_pool=64)
l_inc_4d = inception(l_inc_4c, 'inception_4d', no_1x1=112, no_3x3r=144, no_3x3=288, no_5x5r=32, no_5x5=64,
no_pool=64)
l_inc_4e = inception(l_inc_4d, 'inception_4e', no_1x1=256, no_3x3r=160, no_3x3=320, no_5x5r=32, no_5x5=128,
no_pool=128)
l_pool4 = flip(dnn.MaxPool2DDNNLayer(
flip(l_inc_4e),
pool_size=(3, 3), # pool_size
stride=(2, 2),
pad=1,
name='pool4/3x3_s2'
))
l_inc_5a = inception(l_pool4, 'inception_5a', no_1x1=256, no_3x3r=160, no_3x3=320, no_5x5r=32, no_5x5=128,
no_pool=128)
l_inc_5b = inception(l_inc_5a, 'inception_5b', no_1x1=384, no_3x3r=192, no_3x3=384, no_5x5r=48, no_5x5=128,
no_pool=128)
l_pool5 = flip(dnn.Pool2DDNNLayer(
flip(l_inc_5b),
pool_size=(7, 7), # pool_size
stride=(1, 1),
pad=0,
mode='average',
name='pool5/7x7_s1'
))
l_logit = dnn.Conv2DDNNLayer(
l_pool5,
num_filters=1000,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.linear,
name='prob',
)
l_logit_flat = lasagne.layers.FlattenLayer(l_logit)
l_dense = lasagne.layers.NonlinearityLayer(l_logit_flat, nonlinearity=lasagne.nonlinearities.softmax)
l_out = l_dense
filename = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'googlenet.pkl')
if os.path.exists(filename):
with open(filename, 'r') as f:
nn.layers.set_all_param_values(l_dense, pickle.load(f))
return utils.struct(
input=l_in,
logit=l_logit,
out=l_out,
interesting=l_inc_4c,
)
import healpy as hp
import numpy as np
from healpy import Rotator
import pymaster as nmt
import pandas as pd
from env_config import DATA_PATH
from utils import struct, get_masked_map, tansform_map_and_mask_to_nside
cmb_columns_idx = struct(
I_STOKES=0,
Q_STOKES=1,
U_STOKES=2,
TMASK=3,
PMASK=4,
I_STOKES_INP=5,
Q_STOKES_INP=6,
U_STOKES_INP=7,
TMASK_INP=8,
PMASKINP=9,
)
def get_cmb_temperature_power_spectra(nside):
assert 3 * nside <= 2508
path = os.path.join(DATA_PATH,
'Planck2018/COM_PowerSpect_CMB-TT-full_R3.01.txt')
data = np.loadtxt(path, unpack=False)
data = pd.DataFrame(data, columns=['l', 'Dl', '-dDl', '+dDl'])
def __init__(self):
super(MainWindow, self).__init__()
self.setAcceptDrops(True)
self.loader = None
self.dirty = False
self._selectSlotBlock = False
self._config = config.get_default_config()
self.labelListWidget = LabelListWidget(self)
self.labelListDock = QtWidgets.QDockWidget('Displayed Label List',
self)
self.labelListDock.setWidget(self.labelListWidget)
self.allLabelList = LabelListWidget(self)
self.allLabelListDock = QtWidgets.QDockWidget('Label List', self)
self.allLabelListDock.setWidget(self.allLabelList)
self.colorTableWidget = TaglistWidget(self)
self.colorTableWidget.loadFromJson(self._config['tags'])
self.colorTableDock = QtWidgets.QDockWidget('Color Table', self)
self.colorTableDock.setWidget(self.colorTableWidget)
self.addDockWidget(Qt.RightDockWidgetArea, self.allLabelListDock)
self.addDockWidget(Qt.RightDockWidgetArea, self.colorTableDock)
self.tabifyDockWidget(self.allLabelListDock, self.labelListDock)
self.view = BaseView()
self.setCentralWidget(self.view)
self.statusBar().show()
self.resize(1200, 800)
action = functools.partial(utils.createAction, self)
open_ = action("&Open", self.open, 'open',
self.tr('Open image or label file'))
exit_ = action("&Exit", tip=self.tr('Quit Application'))
openDir_ = action("&Open Dir", self.open, 'open')
createMode_ = action("&Create Polygons",
lambda: self.toggleDrawMode(type.Mode_polygon),
'objects', self.tr('Start drawing polygons'))
createRectangleMode_ = action(
self.tr('Create Rectangle'),
lambda: self.toggleDrawMode(type.Mode_rectangle),
'objects',
self.tr('Start drawing rectangles'),
enabled=False,
)
createCircleMode_ = action(
self.tr('Create Circle'),
lambda: self.toggleDrawMode(type.Mode_circle),
'objects',
self.tr('Start drawing circles'),
enabled=False,
)
createLineMode_ = action(
self.tr('Create Line'),
lambda: self.toggleDrawMode(type.Mode_line),
'objects',
self.tr('Start drawing lines'),
enabled=False,
)
createPointMode_ = action(
self.tr('Create Point'),
lambda: self.toggleDrawMode(type.Mode_point),
'objects',
self.tr('Start drawing points'),
enabled=False,
)
createLineStripMode_ = action(
self.tr('Create LineStrip'),
lambda: self.toggleDrawMode(type.Mode_linestrip),
'objects',
self.tr('Start drawing linestrip. Ctrl+LeftClick ends creation.'),
enabled=False,
)
createBoxMode_ = action(
self.tr('Create Box'),
lambda: self.toggleDrawMode(type.Mode_box),
'objects',
self.tr('Start drawing box.'),
enabled=False,
)
delete_ = action(self.tr('Delete Polygons'),
self.deleteSelectedShape,
'cancel',
self.tr('Delete the selected polygons'),
enabled=False)
deleteAll_ = action(self.tr('Delete All'),
self.deleteSelectedShape,
'delete',
self.tr('Delete all polygons'),
enabled=False)
edit_ = action('&Edit Label',
lambda: self.toggleDrawMode(None),
'edit',
'Modify the label of the selected polygon',
enabled=False)
save_ = action(self.tr('&Save'),
self.saveFile,
'save',
self.tr('Save labels to file'),
enabled=False)
saveAs_ = action(self.tr('&Save As'),
self.saveFileAs,
'save-as',
self.tr('Save labels to a different file'),
enabled=False)
nextTag_ = action(
self.tr('&Next Tag'),
slot=lambda: self.colorTableWidget.selectNext(),
tip=self.tr('Go to a next tag'),
)
prevTag_ = action(
self.tr('&Previous Tag'),
slot=lambda: self.colorTableWidget.selectPrev(),
tip=self.tr('Go to a previous tag'),
)
homeTag_ = action(
self.tr('&Home Tag'),
slot=lambda: self.colorTableWidget.selectHome(),
tip=self.tr('Go to a start tag'),
)
endTag_ = action(
self.tr('&End Tag'),
slot=lambda: self.colorTableWidget.selectEnd(),
tip=self.tr('Go to a end tag'),
)
deleteTag_ = action(
self.tr('&Delete Tag'),
slot=lambda: self.colorTableWidget.deleteSelected())
addTag_ = action(self.tr('&Add Tag'),
slot=lambda: self.colorTableWidget.addTag())
insertTag_ = action(self.tr('&Insert Tag'),
slot=lambda: self.colorTableWidget.insertTag())
fitWindow_ = action(self.tr('&Fit Window'),
slot=self.fitWindow,
icon='fit-window',
tip=self.tr('Zoom follows window size'))
undo_ = action(self.tr('&Undo'),
slot=self.undo,
icon='undo',
tip=self.tr('undo'))
self.zoom_widget = ZoomWidget()
zoom_ = QtWidgets.QWidgetAction(self)
zoom_.setDefaultWidget(self.zoom_widget)
self.actions = utils.struct(
open=open_,
exit=exit_,
openDir=openDir_,
createMode=createMode_,
createRectangleMode=createRectangleMode_,
createCircleMode=createCircleMode_,
createLineMode=createLineMode_,
createPointMode=createPointMode_,
createLineStripMode=createLineStripMode_,
createBoxMode=createBoxMode_,
edit=edit_,
delete=delete_,
save=save_,
saveAs=saveAs_,
nextTag=nextTag_,
prevTag=prevTag_,
homeTag=homeTag_,
endTag=endTag_,
deleteTag=deleteTag_,
fitWindow=fitWindow_,
deleteAll=deleteAll_,
undo=undo_,
# load=load_,
fileMenu=(
open_,
openDir_,
None,
save_,
saveAs_,
None,
# load_,
None,
exit_,
),
editMenu=(
createMode_,
createRectangleMode_,
createCircleMode_,
createLineMode_,
createPointMode_,
createLineStripMode_,
createBoxMode_,
None,
edit_,
undo_,
None,
delete_,
deleteAll_,
),
selectionMenu=(
nextTag_,
prevTag_,
homeTag_,
endTag_,
),
viewMenu=(
self.labelListDock.toggleViewAction(),
self.allLabelListDock.toggleViewAction(),
self.colorTableDock.toggleViewAction(),
None,
fitWindow_,
),
labelListMenu=(delete_, ),
tagListMenu=(
addTag_,
insertTag_,
deleteTag_,
))
self.tools_actions = (
open_,
openDir_,
save_,
saveAs_,
None,
delete_,
deleteAll_,
undo_,
None,
createMode_,
edit_,
None,
zoom_,
fitWindow_,
)
self.toolbar = ToolBar('toolbar')
self.toolbar.setToolButtonStyle(Qt.ToolButtonTextUnderIcon)
self.addToolBar(Qt.TopToolBarArea, self.toolbar)
utils.addActions(self.toolbar, self.tools_actions)
# utils.addActions(self.canvas.menu, self.actions.editMenu)
self.view.addMenu(self.actions.editMenu)
self.labelListWidget.addMenu(self.actions.labelListMenu)
self.allLabelList.addMenu(self.actions.labelListMenu)
self.colorTableWidget.addMenu(self.actions.tagListMenu)
self.addMenu("&File", self.actions.fileMenu)
self.addMenu("&Edit", self.actions.editMenu)
self.addMenu("&Selection", self.actions.selectionMenu)
self.addMenu("&View", self.actions.viewMenu)
# signal
self.zoom_widget.valueChanged.connect(self.zoomChanged)
self.labelListWidget.itemChanged.connect(self.labelItemChanged)
self.labelListWidget.itemDoubleClicked.connect(
lambda: (self.toggleDrawMode(None), self.labelSelectionChanged()))
self.labelListWidget.itemSelectionChanged.connect(
self.labelSelectionChanged)
self.labelListWidget.itemDeleteSelected.connect(
self.deleteSelectedShape)
self.allLabelList.itemSelectionChanged.connect(
self.allLabelListSelected)
self.allLabelList.itemDoubleClicked.connect(self.editLabel)
self.colorTableWidget.itemsDelete.connect(self.tagsDelete)
self.init()
self.initViewSlot()
def vgg19(batch_shape):
"""
Create a vgg19, with the parameters from http://www.robots.ox.ac.uk/~vgg/research/very_deep/
See googlenet.py for the method used to convert these caffe parameters to lasagne parameters.
:param batch_shape: The shape of the input images. This should be of size (N, 3, X>=224, Y>=224). Note flexible
image size, as the last dense layers have been implemented here with convolutional layers.
:return: a struct with the input layer, the logit layer (before the final softmax) and the output layer.
"""
l_in = nn.layers.InputLayer(shape=batch_shape)
l = l_in
l = conv3(l, num_filters=64)
l = conv3(l, num_filters=64)
l = max_pool(l)
l = conv3(l, num_filters=128)
l = conv3(l, num_filters=128)
l = max_pool(l)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = conv3(l, num_filters=256)
l = max_pool(l)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = max_pool(l)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = conv3(l, num_filters=512)
l = max_pool(l)
l = dnn.Conv2DDNNLayer(l,
num_filters=4096,
stride=(1, 1),
border_mode="valid",
filter_size=(7, 7))
l = dnn.Conv2DDNNLayer(l,
num_filters=4096,
stride=(1, 1),
border_mode="same",
filter_size=(1, 1))
l_logit = dnn.Conv2DDNNLayer(l,
num_filters=1000,
stride=(1, 1),
border_mode="same",
filter_size=(1, 1),
nonlinearity=None)
l_logit_flat = nn.layers.FlattenLayer(l_logit)
l_dense = nn.layers.NonlinearityLayer(
l_logit_flat, nonlinearity=nn.nonlinearities.softmax)
filename = os.path.join(os.path.dirname(os.path.abspath(__file__)),
'vgg19.pkl')
if os.path.exists(filename):
with open(filename, 'r') as f:
nn.layers.set_all_param_values(l_dense, pickle.load(f))
return utils.struct(input=l_in, out=l_dense, logit=l_logit)
def build_model(batch_size=batch_size, seq_length=seq_length):
l_in = nn.layers.InputLayer(shape=(batch_size, seq_length, n_channels)+im_shape)
l = l_in
_,s,_,h,w = l.get_output_shape()
l = nn.layers.ReshapeLayer(l, (batch_size*seq_length,n_channels)+im_shape)
n = 16
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = reshape_bchw_to_bcs(l, s)
l = conv1d_layer(l, num_filters=n, filter_length=3)
l = reshape_bcs_to_bchw(l, h, w)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = reshape_bchw_to_bcs(l, s)
l = conv1d_layer(l, num_filters=n, filter_length=3)
l = reshape_bcs_to_bchw(l, h, w)
l = maxpool(l, s, ds=(2,2,2))
n = 32
_,s,_,h,w = l.get_output_shape()
l = reshape_bschw_to_bchw(l)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = reshape_bchw_to_bcs(l,s)
l = conv1d_layer(l, num_filters=n, filter_length=3)
l = reshape_bcs_to_bchw(l, h, w)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = reshape_bchw_to_bcs(l, s)
l = conv1d_layer(l, num_filters=n, filter_length=3)
l = reshape_bcs_to_bchw(l, h, w)
l = maxpool(l, s, ds=(2,2,2))
n = 64
_,s,_,h,w = l.get_output_shape()
l = reshape_bschw_to_bchw(l)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = reshape_bchw_to_bcs(l,s)
l = conv1d_layer(l, num_filters=n, filter_length=3)
l = reshape_bcs_to_bchw(l, h, w)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = reshape_bchw_to_bcs(l, s)
l = conv1d_layer(l, num_filters=n, filter_length=3)
l = reshape_bcs_to_bchw(l, h, w)
l = maxpool(l, s, ds=(2,2,2))
n = 128
_,s,_,h,w = l.get_output_shape()
l = reshape_bschw_to_bchw(l)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = reshape_bchw_to_bcs(l,s)
l = conv1d_layer(l, num_filters=n, filter_length=3)
l = reshape_bcs_to_bchw(l, h, w)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = reshape_bchw_to_bcs(l, s)
l = conv1d_layer(l, num_filters=n, filter_length=3)
l = reshape_bcs_to_bchw(l, h, w)
l = maxpool(l, s, ds=(2,2,2))
l = nn.layers.DropoutLayer(l, p=p_shared)
l = dense_layer(l, num_units=2048)
l = nn.layers.DropoutLayer(l, p=p_shared)
l = dense_layer(l, num_units=2048)
l = nn.layers.DropoutLayer(l, p=p_shared)
l_out = nn.layers.DenseLayer(l, num_units=n_classes,
nonlinearity=nn.nonlinearities.softmax)
return utils.struct(
input=l_in,
out=l_out)
from utils import struct, count, any#, severs
tot=defaultdict(lambda:0)
hit=defaultdict(lambda:0)
cats=20
def record(sev, truth):
post = 1.0 / (1 + 1/sev) # prior=1/2
rpost = round(post*cats)
tot[rpost] += 1
if truth:
hit[rpost] += 1
for i in range(1):
net=struct()
net.a = random()
net.b_given_a=[random(), random()]
c_given_b = [random(), random()]
net.c_given_ab = [c_given_b] * 2
net.d_given_a = [random(), random()]
data = simulate(net, 1000)
cnt = count(zip(*data))
if any(cnt, lambda(x):x==0):
print 'net: %s' % net
print 'skipping %s' % cnt
continue
bf = has_common_cause_bf(data[0:3])
record(bf, False)
bf = has_common_cause_bf(data[1:4])
record(bf, True)
import traceback
from collections import deque
from functools import reduce
import pp
from lex import lex
from stack import Stack
from utils import struct
# TODO: capture stdin, stdout, stderr
Record = struct('cmd', 'args', 'result')
class Stark:
def __init__(self, hist_length=1000):
self.hist = deque(maxlen=hist_length)
self.stack = Stack({
name[4:].replace('_', '-'): getattr(self, name)
for name in dir(self)
if name.startswith('cmd_')
})
def cmd_record(self, *args, **kwargs):
self.hist.append(Record(*args, **kwargs))
def cmd_lex(self, code):
return lex(code)
def googlenet(batch_shape):
"""
Create a googlenet, with the parameters from https://github.com/BVLC/caffe/tree/master/models/bvlc_googlenet
:param batch_shape: The shape of the input images. This should be of size (N, 3, 224, 224)
:return: a struct with the input layer, the logit layer (before the final softmax) and the output layer.
"""
l_in = lasagne.layers.InputLayer(
shape=batch_shape,
name='input',
)
l_conv1 = dnn.Conv2DDNNLayer(
l_in,
num_filters=64,
filter_size=(7, 7),
pad=3,
stride=(2, 2),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv1/7x7_s2',
)
l_pool1 = flip(
dnn.MaxPool2DDNNLayer(
flip(l_conv1),
pool_size=(3, 3), # pool_size
stride=(2, 2),
pad=(1, 1),
name='pool1/3x3_s2'))
lrn = LocalResponseNormalization2DLayer(
l_pool1,
alpha=0.0001 / 5,
beta=0.75,
k=1,
n=5,
name='pool1/norm1',
)
l_conv2 = dnn.Conv2DDNNLayer(
lrn,
num_filters=64,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv2/3x3_reduce',
)
l_conv2b = dnn.Conv2DDNNLayer(
l_conv2,
num_filters=192,
filter_size=(3, 3),
pad=1,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name='conv2/3x3',
)
lrn2 = LocalResponseNormalization2DLayer(
l_conv2b,
alpha=0.0001 / 5,
beta=0.75,
k=1,
n=5,
name='conv2/norm2',
)
l_pool2 = flip(
dnn.MaxPool2DDNNLayer(
flip(lrn2),
pool_size=(3, 3), # pool_size
stride=(2, 2),
pad=(1, 1),
name='pool2/3x3_s2'))
def inception(layer,
name,
no_1x1=64,
no_3x3r=96,
no_3x3=128,
no_5x5r=16,
no_5x5=32,
no_pool=32):
l_conv_inc = dnn.Conv2DDNNLayer(
layer,
num_filters=no_1x1,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/1x1',
)
l_conv_inc2 = dnn.Conv2DDNNLayer(
layer,
num_filters=no_3x3r,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/3x3_reduce',
)
l_conv_inc2b = dnn.Conv2DDNNLayer(
l_conv_inc2,
num_filters=no_3x3,
filter_size=(3, 3),
pad=1,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/3x3',
)
l_conv_inc2c = dnn.Conv2DDNNLayer(
layer,
num_filters=no_5x5r,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/5x5_reduce',
)
l_conv_inc2d = dnn.Conv2DDNNLayer(
l_conv_inc2c,
num_filters=no_5x5,
filter_size=(5, 5),
pad=2,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/5x5',
)
l_pool2 = flip(
dnn.MaxPool2DDNNLayer(
flip(layer),
pool_size=(3, 3), # pool_size
stride=(1, 1),
pad=(1, 1),
name=name + '/pool'))
l_conv_inc2e = dnn.Conv2DDNNLayer(
l_pool2,
num_filters=no_pool,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.rectify,
name=name + '/pool_proj',
)
l_inc_out = nn.layers.concat(
[l_conv_inc, l_conv_inc2b, l_conv_inc2d, l_conv_inc2e])
return l_inc_out
l_inc_3a = inception(l_pool2,
'inception_3a',
no_1x1=64,
no_3x3r=96,
no_3x3=128,
no_5x5r=16,
no_5x5=32,
no_pool=32)
l_inc_3b = inception(l_inc_3a,
'inception_3b',
no_1x1=128,
no_3x3r=128,
no_3x3=192,
no_5x5r=32,
no_5x5=96,
no_pool=64)
l_pool3 = flip(
dnn.MaxPool2DDNNLayer(
flip(l_inc_3b),
pool_size=(3, 3), # pool_size
stride=(2, 2),
pad=1,
name='pool3/3x3_s2'))
l_inc_4a = inception(l_pool3,
'inception_4a',
no_1x1=192,
no_3x3r=96,
no_3x3=208,
no_5x5r=16,
no_5x5=48,
no_pool=64)
l_inc_4b = inception(l_inc_4a,
'inception_4b',
no_1x1=160,
no_3x3r=112,
no_3x3=224,
no_5x5r=24,
no_5x5=64,
no_pool=64)
l_inc_4c = inception(l_inc_4b,
'inception_4c',
no_1x1=128,
no_3x3r=128,
no_3x3=256,
no_5x5r=24,
no_5x5=64,
no_pool=64)
l_inc_4d = inception(l_inc_4c,
'inception_4d',
no_1x1=112,
no_3x3r=144,
no_3x3=288,
no_5x5r=32,
no_5x5=64,
no_pool=64)
l_inc_4e = inception(l_inc_4d,
'inception_4e',
no_1x1=256,
no_3x3r=160,
no_3x3=320,
no_5x5r=32,
no_5x5=128,
no_pool=128)
l_pool4 = flip(
dnn.MaxPool2DDNNLayer(
flip(l_inc_4e),
pool_size=(3, 3), # pool_size
stride=(2, 2),
pad=1,
name='pool4/3x3_s2'))
l_inc_5a = inception(l_pool4,
'inception_5a',
no_1x1=256,
no_3x3r=160,
no_3x3=320,
no_5x5r=32,
no_5x5=128,
no_pool=128)
l_inc_5b = inception(l_inc_5a,
'inception_5b',
no_1x1=384,
no_3x3r=192,
no_3x3=384,
no_5x5r=48,
no_5x5=128,
no_pool=128)
l_pool5 = flip(
dnn.Pool2DDNNLayer(
flip(l_inc_5b),
pool_size=(7, 7), # pool_size
stride=(1, 1),
pad=0,
mode='average',
name='pool5/7x7_s1'))
l_logit = dnn.Conv2DDNNLayer(
l_pool5,
num_filters=1000,
filter_size=(1, 1),
pad=0,
stride=(1, 1),
nonlinearity=lasagne.nonlinearities.linear,
name='prob',
)
l_logit_flat = lasagne.layers.FlattenLayer(l_logit)
l_dense = lasagne.layers.NonlinearityLayer(
l_logit_flat, nonlinearity=lasagne.nonlinearities.softmax)
l_out = l_dense
filename = os.path.join(os.path.dirname(os.path.abspath(__file__)),
'googlenet.pkl')
if os.path.exists(filename):
with open(filename, 'r') as f:
nn.layers.set_all_param_values(l_dense, pickle.load(f))
return utils.struct(
input=l_in,
logit=l_logit,
out=l_out,
interesting=l_inc_4c,
)
def build_model(batch_size=batch_size, seq_length=seq_length):
l_in = nn.layers.InputLayer(shape=(batch_size, seq_length, n_channels)+im_shape)
l = l_in
l = nn.layers.ReshapeLayer(l, (batch_size*seq_length,n_channels)+im_shape)
n = 16
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = conv1d_layer(l, num_filters=n, filter_size=3)
l = conv1d_layer(l, num_filters=n, filter_size=3)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = maxpool(l, pool_size=(2,2))
n = 32
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = conv1d_layer(l, num_filters=n, filter_size=3)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = conv1d_layer(l, num_filters=n, filter_size=3)
l = maxpool(l, pool_size=(2,2))
n = 64
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = conv1d_layer(l, num_filters=n, filter_size=3)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = conv1d_layer(l, num_filters=n, filter_size=3)
l = maxpool(l, pool_size=(2,2))
n = 128
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = conv1d_layer(l, num_filters=n, filter_size=3)
l = conv_layer(l, num_filters=n, filter_size=(3,3))
l = conv1d_layer(l, num_filters=n, filter_size=3)
l = maxpool(l, pool_size=(2,2))
l = nn.layers.ReshapeLayer(l, (batch_size, seq_length, -1))
l = nn.layers.DropoutLayer(l, p=p_shared)
n_states = 512
l_fwd = lstm_layer(l, n_states)
l_bck = lstm_layer(l, n_states, backwards=True)
l = nn.layers.ElemwiseSumLayer([l_fwd, l_bck])
l = nn.layers.DropoutLayer(l, p=p_shared)
l = nn.layers.ReshapeLayer(l, (batch_size*seq_length, n_states))
l_out = nn.layers.DenseLayer(l, num_units=n_classes,
nonlinearity=nn.nonlinearities.softmax)
l_out_reshape = nn.layers.ReshapeLayer(l_out, (batch_size, seq_length, n_classes))
return utils.struct(
input=l_in,
out=l_out,
out_reshape=l_out_reshape)
def getTagObj(self, id=0):
obj = utils.struct(id=id,
color=self.color_edit.getColor().getRgb(),
desc=self.line_edit.text())
return obj
