def load_test(CVsplit):
print "loadTest started!"
xb_test = utils.load_gz('data/numpy/test/Btst%s.npy.gz' % CVsplit).astype('float32')
tb_test = np.zeros((xb_test.shape[0],1), dtype='float32')
xs_test = utils.load_gz('data/numpy/test/Stst%s.npy.gz' % CVsplit).astype('float32')
ts_test = np.ones((xs_test.shape[0],1), dtype='float32')
return xb_test, tb_test, xs_test, ts_test
def load_test(CVsplit):
print "loadTest started!"
xb_test = utils.load_gz('data/numpy/test/Btst%s.npy.gz' %
CVsplit).astype('float32')
tb_test = np.zeros((xb_test.shape[0], 1), dtype='float32')
xs_test = utils.load_gz('data/numpy/test/Stst%s.npy.gz' %
CVsplit).astype('float32')
ts_test = np.ones((xs_test.shape[0], 1), dtype='float32')
return xb_test, tb_test, xs_test, ts_test
def load_data(CVsplit):
print "loadData started!"
xb_train = utils.load_gz('data/numpy/train/Btrn%s.npy.gz' % CVsplit).astype('float32')
tb_train = np.zeros((xb_train.shape[0],1), dtype='float32')
xs_train = utils.load_gz('data/numpy/train/Strn%s.npy.gz' % CVsplit).astype('float32')
ts_train = np.ones((xs_train.shape[0],1), dtype='float32')
xb_valid = utils.load_gz('data/numpy/train/Bval%s.npy.gz' % CVsplit).astype('float32')
tb_valid = np.zeros((xb_valid.shape[0],1), dtype='float32')
xs_valid = utils.load_gz('data/numpy/train/Sval%s.npy.gz' % CVsplit).astype('float32')
ts_valid = np.ones((xs_valid.shape[0],1), dtype='float32')
return xb_train, xb_valid, tb_train, tb_valid, xs_train, xs_valid, ts_train, ts_valid
def load_test(CVsplit):
if not os.path.isfile(TEST_PATH):
print("Downloading and extracting test ...")
subprocess.call("bash download_test.sh", shell=True)
subprocess.call("python create_data.py test", shell=True)
else:
print "Test already downloaded ..."
xb_test = utils.load_gz('data/numpy/test/Btst%s.npy.gz' % CVsplit).astype('float32')
tb_test = np.zeros((xb_test.shape[0]), dtype='float32')
xs_test = utils.load_gz('data/numpy/test/Stst%s.npy.gz' % CVsplit).astype('float32')
ts_test = np.ones((xs_test.shape[0]), dtype='float32')
return xb_test, tb_test, xs_test, ts_test
def load_test(CVsplit):
if not os.path.isfile(TEST_PATH):
print("Downloading and extracting test ...")
subprocess.call("bash download_test.sh", shell=True)
subprocess.call("python create_data.py test", shell=True)
else:
print "Test already downloaded ..."
xb_test = utils.load_gz('data/numpy/test/Btst%s.npy.gz' %
CVsplit).astype('float32')
tb_test = np.zeros((xb_test.shape[0]), dtype='float32')
xs_test = utils.load_gz('data/numpy/test/Stst%s.npy.gz' %
CVsplit).astype('float32')
ts_test = np.ones((xs_test.shape[0]), dtype='float32')
return xb_test, tb_test, xs_test, ts_test
def load_data(CVsplit):
print "loadData started!"
xb_train = utils.load_gz('data/numpy/train/Btrn%s.npy.gz' %
CVsplit).astype('float32')
tb_train = np.zeros((xb_train.shape[0], 1), dtype='float32')
xs_train = utils.load_gz('data/numpy/train/Strn%s.npy.gz' %
CVsplit).astype('float32')
ts_train = np.ones((xs_train.shape[0], 1), dtype='float32')
xb_valid = utils.load_gz('data/numpy/train/Bval%s.npy.gz' %
CVsplit).astype('float32')
tb_valid = np.zeros((xb_valid.shape[0], 1), dtype='float32')
xs_valid = utils.load_gz('data/numpy/train/Sval%s.npy.gz' %
CVsplit).astype('float32')
ts_valid = np.ones((xs_valid.shape[0], 1), dtype='float32')
return xb_train, xb_valid, tb_train, tb_valid, xs_train, xs_valid, ts_train, ts_valid
def get_train():
print('[Info] Build training and validation dataset ...')
X_in = load_gz(TRAIN_PATH)
X = np.reshape(X_in, (5534, 700, 57))
del X_in
X = X[:, :, :]
X_total, labels_total = get_instances_from_file(X)
# getting meta
num_seqs = np.size(X_total, 0)
seq_names = np.arange(0, num_seqs)
print('[Info] Num of training and validation dataset: ', num_seqs)
#boundary = 5278
boundary = int(0.9 * num_seqs)
X_train = X_total[seq_names[0:boundary]]
labels_train = labels_total[seq_names[0:boundary]]
X_valid = X_total[seq_names[boundary:num_seqs]]
labels_valid = labels_total[seq_names[boundary:num_seqs]]
return X_train.tolist(), labels_train.tolist(), X_valid.tolist(
), labels_valid.tolist()
def load_train(CVsplit):
if not os.path.isfile(TRAIN_PATH):
print("Downloading and extracting train ...")
subprocess.call("bash download_train.sh", shell=True)
subprocess.call("python create_data.py train", shell=True)
else:
print "Train already downloaded ..."
xb_train = utils.load_gz('data/numpy/train/Btrn%s.npy.gz' % CVsplit).astype('float32')
tb_train = np.zeros((xb_train.shape[0]), dtype='float32')
xs_train = utils.load_gz('data/numpy/train/Strn%s.npy.gz' % CVsplit).astype('float32')
ts_train = np.ones((xs_train.shape[0]), dtype='float32')
xb_valid = utils.load_gz('data/numpy/train/Bval%s.npy.gz' % CVsplit).astype('float32')
tb_valid = np.zeros((xb_valid.shape[0]), dtype='float32')
xs_valid = utils.load_gz('data/numpy/train/Sval%s.npy.gz' % CVsplit).astype('float32')
ts_valid = np.ones((xs_valid.shape[0]), dtype='float32')
return xb_train, xb_valid, tb_train, tb_valid, xs_train, xs_valid, ts_train, ts_valid
def get_train(path=TRAIN_PATH):
if not os.path.isfile(path):
print("Train path is not downloaded ...")
subprocess.call("./download_train.sh", shell=True)
else:
print("Train path is downloaded ...")
print("Loading train data ...")
X_in = utils.load_gz(path)
X = np.reshape(X_in, (5534, 700, 57))
del X_in
X = X[:, :, :]
labels = X[:, :, 22:30]
mask = X[:, :, 30] * -1 + 1
a = np.arange(0, 21)
b = np.arange(35, 56)
c = np.hstack((a, b))
X = X[:, :, c]
# getting meta
num_seqs = np.size(X, 0)
seqlen = np.size(X, 1)
d = np.size(X, 2)
num_classes = 8
#### REMAKING LABELS ####
X = X.astype(theano.config.floatX)
mask = mask.astype(theano.config.floatX)
# Dummy -> concat
vals = np.arange(0, 8)
labels_new = np.zeros((num_seqs, seqlen))
for i in range(np.size(labels, axis=0)):
labels_new[i, :] = np.dot(labels[i, :, :], vals)
labels_new = labels_new.astype('int32')
labels = labels_new
print("Loading splits ...")
##### SPLITS #####
# getting splits (cannot run before splits are made)
# split = np.load("data/split.pkl")
seq_names = np.arange(0, num_seqs)
# np.random.shuffle(seq_names)
X_train = X[seq_names[0:5278]]
X_valid = X[seq_names[5278:5534]]
labels_train = labels[seq_names[0:5278]]
labels_valid = labels[seq_names[5278:5534]]
mask_train = mask[seq_names[0:5278]]
mask_valid = mask[seq_names[5278:5534]]
num_seq_train = np.size(X_train, 0)
num_seq_valid = np.size(X_valid, 0)
return X_train, X_valid, labels_train, labels_valid, mask_train, \
mask_valid, num_seq_train
def get_test():
print('[Info] Build testing dataset ...')
X_test_in = load_gz(TEST_PATH)
X_test = np.reshape(X_test_in, (514, 700, 57))
del X_test_in
X_test = X_test[:, :, :]
X_test, labels_test = get_instances_from_file(X_test)
print('[Info] Num of test dataset: ', np.size(X_test, 0))
return X_test.tolist(), labels_test.tolist()
def load_train(CVsplit):
if not os.path.isfile(TRAIN_PATH):
print("Downloading and extracting train ...")
subprocess.call("bash download_train.sh", shell=True)
subprocess.call("python create_data.py train", shell=True)
else:
print "Train already downloaded ..."
xb_train = utils.load_gz('data/numpy/train/Btrn%s.npy.gz' %
CVsplit).astype('float32')
tb_train = np.zeros((xb_train.shape[0]), dtype='float32')
xs_train = utils.load_gz('data/numpy/train/Strn%s.npy.gz' %
CVsplit).astype('float32')
ts_train = np.ones((xs_train.shape[0]), dtype='float32')
xb_valid = utils.load_gz('data/numpy/train/Bval%s.npy.gz' %
CVsplit).astype('float32')
tb_valid = np.zeros((xb_valid.shape[0]), dtype='float32')
xs_valid = utils.load_gz('data/numpy/train/Sval%s.npy.gz' %
CVsplit).astype('float32')
ts_valid = np.ones((xs_valid.shape[0]), dtype='float32')
return xb_train, xb_valid, tb_train, tb_valid, xs_train, xs_valid, ts_train, ts_valid
def get_train():
if not os.path.isfile(TRAIN_PATH):
print("No Training Data Available ...")
else:
print("Training Data is Available ...")
print("Loading train data ...")
x_in = utils.load_gz(TRAIN_PATH)
x = np.reshape(x_in, (5534, 700, 57))
del x_in
x = x[:, :, :]
labels = x[:, :, 22:30]
mask = x[:, :, 30] * -1 + 1
amino_acid_residues = np.arange(0, 21)
sequence_profile = np.arange(35, 56)
horizontal_stack = np.hstack((amino_acid_residues, sequence_profile))
x = x[:, :, horizontal_stack]
print('x: ', x)
# getting meta #
num_seqs_row = np.size(x, 0)
seqlen_column = np.size(x, 1)
# REMAKING LABELS #
x = x.astype(theano.config.floatX)
mask = mask.astype(theano.config.floatX)
# Dummy -> concat
vals = np.arange(0, 8)
labels_new = np.zeros((num_seqs_row, seqlen_column))
for i in xrange(np.size(labels, axis=0)):
labels_new[i, :] = np.dot(labels[i, :, :], vals)
labels_new = labels_new.astype('int32')
labels = labels_new
print("labels: ", labels)
print("Loading splits ...")
# SPLITS #
seq_names = np.arange(0, num_seqs_row)
x_train = x[seq_names[0:5278]]
x_valid = x[seq_names[5278:5534]]
labels_train = labels[seq_names[0:5278]]
labels_valid = labels[seq_names[5278:5534]]
mask_train = mask[seq_names[0:5278]]
mask_valid = mask[seq_names[5278:5534]]
num_seq_train = np.size(x_train, 0)
return x_train, x_valid, labels_train, labels_valid, mask_train, mask_valid, num_seq_train
def get_test(seq_len=None):
if not os.path.isfile(TEST_PATH):
subprocess.call("./download_test.sh", shell=True)
print("Loading test data ...")
X_test_in = utils.load_gz(TEST_PATH)
X_test = np.reshape(X_test_in, (514, 700, 57))
del X_test_in
X_test = X_test[:, :, :].astype("float32")
labels_test = X_test[:, :, 22:30].astype('int32')
mask_test = X_test[:, :, 30].astype("float32") * -1 + 1
a = np.arange(0, 21)
b = np.arange(35, 56)
c = np.hstack((a, b))
X_test = X_test[:, :, c]
# getting meta
seqlen = np.size(X_test, 1)
d = np.size(X_test, 2)
num_classes = 8
num_seq_test = np.size(X_test, 0)
del a, b, c
## DUMMY -> CONCAT ##
vals = np.arange(0, 8)
labels_new = np.zeros((num_seq_test, seqlen))
for i in xrange(np.size(labels_test, axis=0)):
labels_new[i, :] = np.dot(labels_test[i, :, :], vals)
labels_new = labels_new.astype('int32')
labels_test = labels_new
### ADDING BATCH PADDING ###
X_add = np.zeros((126, seqlen, d))
label_add = np.zeros((126, seqlen))
mask_add = np.zeros((126, seqlen))
#
X_test = np.concatenate((X_test, X_add), axis=0).astype("float32")
labels_test = np.concatenate((labels_test, label_add),
axis=0).astype('int32')
mask_test = np.concatenate((mask_test, mask_add), axis=0).astype("float32")
if seq_len is not None:
X_test = X_test[:, :seq_len]
labels_test = labels_test[:, :seq_len]
mask_test = mask_test[:, :seq_len]
len_test = np.sum(mask_test, axis=1)
len_test[-126:] = np.ones((126, ), dtype='int32')
return X_test, mask_test, labels_test, num_seq_test, len_test
def make_dataset(path):
data = load_gz(path)
data = data.reshape(-1, 700, 57)
idx = np.append(np.arange(21), np.arange(35, 56))
X = data[:, :, idx]
X = X.transpose(0, 2, 1)
X = X.astype('float32')
y = data[:, :, 22:30]
y = np.array([np.dot(yi, np.arange(8)) for yi in y])
y = y.astype('float32')
mask = data[:, :, 30] * -1 + 1
seq_len = mask.sum(axis=1)
seq_len = seq_len.astype('float32')
return X, y, seq_len
def get_test():
if not os.path.isfile(TEST_PATH):
print("Test Data Unavailable")
print("Loading test data ...")
x_test_in = utils.load_gz(TEST_PATH)
x_test = np.reshape(x_test_in, (514, 700, 57))
del x_test_in
x_test = x_test[:, :, :].astype(theano.config.floatX)
labels_test = x_test[:, :, 22:30].astype('int32')
mask_test = x_test[:, :, 30].astype(theano.config.floatX) * -1 + 1
a = np.arange(0, 21)
b = np.arange(35, 56)
c = np.hstack((a, b))
x_test = x_test[:, :, c]
# getting meta
seqlen = np.size(x_test, 1)
d = np.size(x_test, 2)
num_classes = 8
num_seq_test = np.size(x_test, 0)
del a, b, c
# DUMMY -> CONCAT #
vals = np.arange(0, 8)
labels_new = np.zeros((num_seq_test, seqlen))
for i in xrange(np.size(labels_test, axis=0)):
labels_new[i, :] = np.dot(labels_test[i, :, :], vals)
labels_new = labels_new.astype('int32')
labels_test = labels_new
# ADDING BATCH PADDING #
x_add = np.zeros((126, seqlen, d))
label_add = np.zeros((126, seqlen))
mask_add = np.zeros((126, seqlen))
x_test = np.concatenate((x_test, x_add),
axis=0).astype(theano.config.floatX)
labels_test = np.concatenate((labels_test, label_add),
axis=0).astype('int32')
mask_test = np.concatenate((mask_test, mask_add),
axis=0).astype(theano.config.floatX)
return x_test, mask_test, labels_test, num_seq_test
def load_data(CVsplit):
print "loadData started!"
if(len(glob.glob('./data/csv/*'))!=len(glob.glob('./data/numpy/*'))):
print "converting data ..."
convert_data();
xb_train = utils.load_gz('data/numpy/Btrn%s.npy.gz' % CVsplit).astype('float32')
tb_train = np.zeros((xb_train.shape[0],1), dtype='float32')
xs_train = utils.load_gz('data/numpy/Strn%s.npy.gz' % CVsplit).astype('float32')
ts_train = np.ones((xs_train.shape[0],1), dtype='float32')
xb_test = utils.load_gz('data/numpy/Btst%s.npy.gz' % CVsplit).astype('float32')
tb_test = np.zeros((xb_test.shape[0],1), dtype='float32')
xs_test = utils.load_gz('data/numpy/Stst%s.npy.gz' % CVsplit).astype('float32')
ts_test = np.ones((xs_test.shape[0],1), dtype='float32')
xb_valid = utils.load_gz('data/numpy/Bval%s.npy.gz' % CVsplit).astype('float32')
tb_valid = np.zeros((xb_valid.shape[0],1), dtype='float32')
xs_valid = utils.load_gz('data/numpy/Sval%s.npy.gz' % CVsplit).astype('float32')
ts_valid = np.ones((xs_valid.shape[0],1), dtype='float32')
return xb_train, xb_valid, xb_test, tb_train, tb_valid, tb_test, xs_train, xs_valid, xs_test, ts_train, ts_valid, ts_test
def _prepare_data(self):
if not os.path.exists("data/train.npy.gz"):
self._load_data()
self._preprocess_data()
else:
self.samples = utils.load_gz("data/train.npy.gz")
"""
Created on Wed Jun 17 12:00:27 2015
@author: alexander
"""
import numpy as np
import cPickle as pickle
import sklearn
import sklearn.cross_validation
import theano
import utils
## From data
print("Loading train data ...")
X_in = utils.load_gz('data/cullpdb+profile_6133_filtered.npy.gz')
X = np.reshape(X_in, (5534, 700, 57))
del X_in
X = X[:, 0:100, :]
labels = X[:, :, 22:30]
mask = X[:, :, 30]
a = np.arange(0, 21)
b = np.arange(35, 56)
c = np.hstack((a, b))
X = X[:, :, c]
# getting meta
num_seqs = np.size(X, 0)
seqlen = np.size(X, 1)
d = np.size(X, 2)
paths_train = glob.glob("data/train/*/*")
paths_train.sort()
paths_test = glob.glob("data/test/*")
paths_test.sort()
paths = {
'train': paths_train,
'test': paths_test,
}
# labels_train = np.zeros(len(paths['train']), dtype='int32')
# for k, path in enumerate(paths['train']):
# class_name = os.path.basename(os.path.dirname(path))
# labels_train[k] = class_names.index(class_name)
labels_train = utils.load_gz("data/labels_train.npy.gz")
default_augmentation_params = {
'zoom_range': (1 / 1.1, 1.1),
'rotation_range': (0, 360),
'shear_range': (0, 0),
'translation_range': (-4, 4),
'do_flip': True,
'allow_stretch': False,
}
no_augmentation_params = {
'zoom_range': (1.0, 1.0),
'rotation_range': (0, 0),
'shear_range': (0, 0),
'translation_range': (0, 0),
"""
Created on Wed Jun 17 12:00:27 2015
@author: alexander
"""
import numpy as np
import cPickle as pickle
import sklearn
import sklearn.cross_validation
import theano
import utils
## From data
print("Loading train data ...")
X_in = utils.load_gz('data/cullpdb+profile_6133_filtered.npy.gz')
X = np.reshape(X_in,(5534,700,57))
del X_in
X = X[:,0:100,:]
labels = X[:,:,22:30]
mask = X[:,:,30]
a = np.arange(0,21)
b = np.arange(35,56)
c = np.hstack((a,b))
X = X[:,:,c]
# getting meta
num_seqs = np.size(X,0)
seqlen = np.size(X,1)
import numpy as np
import theano
import utils
import load_protvec
##### TRAIN DATA #####
print("Loading train data ...")
protein_vector_file = "data/protVec_100d_3grams_clean.csv"
addProtVec = False
X_in = utils.load_gz("data/cullpdb+profile_6133_filtered.npy.gz")
X = np.reshape(X_in, (5534, 700, 57))
del X_in
X = X[:, :, :]
labels = X[:, :, 22:30]
mask = X[:, :, 30] * -1 + 1
a = np.arange(0, 21)
b = np.arange(35, 56)
c = np.hstack((a, b))
X = X[:, :, c]
# If using ProtVec
# http://arxiv.org/abs/1503.05140
if addProtVec:
ProtVec_train = load_protvec.load_protvec_encoding(X, mask, protein_vector_file, protvec_dim=100)
newX = np.zeros((X.shape[0], X.shape[1], X.shape[2] + ProtVec_train.shape[2]))
newX[:, :, : X.shape[2]] = X
newX[:, :, X.shape[2] :] = ProtVec_train
X = newX
paths_train.sort()
paths_test = glob.glob("data/test/*")
paths_test.sort()
paths = {
'train': paths_train,
'test': paths_test,
}
# labels_train = np.zeros(len(paths['train']), dtype='int32')
# for k, path in enumerate(paths['train']):
# class_name = os.path.basename(os.path.dirname(path))
# labels_train[k] = class_names.index(class_name)
labels_train = utils.load_gz("data/labels_train.npy.gz")
default_augmentation_params = {
'zoom_range': (1 / 1.1, 1.1),
'rotation_range': (0, 360),
'shear_range': (0, 0),
'translation_range': (-4, 4),
'do_flip': True,
'allow_stretch': False,
}
no_augmentation_params = {
'zoom_range': (1.0, 1.0),
'rotation_range': (0, 0),
'shear_range': (0, 0),
def load(subset='train'):
"""
Load all images into memory for faster processing
"""
return utils.load_gz("data/images_%s.npy.gz" % subset)
def load(subset='train'):
"""
Load all images into memory for faster processing
"""
return utils.load_gz("data/images_%s.npy.gz" % subset)
import numpy as np
import theano
import utils
##### TRAIN DATA #####
print("Loading train data ...")
X_in = utils.load_gz("data/cullpdb+profile_6133_filtered.npy.gz")
X = np.reshape(X_in, (5534, 700, 57))
del X_in
X = X[:, :, :]
labels = X[:, :, 22:30]
mask = X[:, :, 30] * -1 + 1
a = np.arange(0, 21)
b = np.arange(35, 56)
c = np.hstack((a, b))
X = X[:, :, c]
# If using ProtVec
# http://arxiv.org/abs/1503.05140
addProtVec = True
if addProtVec:
ProtVec = utils.load_gz("data/X_train_protvec.npy")
newX = np.zeros((X.shape[0], X.shape[1], X.shape[2] + ProtVec.shape[2]))
newX[:, :, : X.shape[2]] = X
newX[:, :, X.shape[2] :] = ProtVec
X = newX
del newX
# getting meta
class_names = [os.path.basename(d) for d in directories]
class_names.sort()
num_classes = len(class_names)
DEFAULT_VALIDATION_SPLIT = './data/validation_split_v1.pkl'
paths_train = glob.glob("data/train/*/*")
paths_train.sort()
paths_test = glob.glob("data/test/*")
paths_test.sort()
paths = {'train': paths_train, 'test': paths_test}
labels_train_path = 'data/labels_train.npy.gz'
labels_train = utils.load_gz(labels_train_path)
class LoadMethod(object):
def __init__(self, paths, labels=None):
self.paths = paths
self.labels = labels
def __call__(self, idx):
if self.labels is not None:
return skimage.io.imread(self.paths[idx], as_grey=True).astype(
'float32'), self.labels[idx].astype('int32')
else:
return skimage.io.imread(self.paths[idx],
as_grey=True).astype('float32')