def convert_orig_to_np():
from pylearn2.datasets.filetensor import read
import gzip
import cPickle
# Load data
path_orig = os.environ['ML_DATA_PATH']+'/norb_orig/'
prefix_train = path_orig+'smallnorb-5x46789x9x18x6x2x96x96-training-'
train_cat = read(gzip.open(prefix_train+'cat.mat.gz'))
train_dat = read(gzip.open(prefix_train+'dat.mat.gz'))
train_info = read(gzip.open(prefix_train+'info.mat.gz'))
prefix_test = path_orig+'smallnorb-5x01235x9x18x6x2x96x96-testing-'
test_cat = read(gzip.open(prefix_test+'cat.mat.gz'))
test_dat = read(gzip.open(prefix_test+'dat.mat.gz'))
test_info = read(gzip.open(prefix_test+'info.mat.gz'))
# Save originals matrices to file
files = (('train_cat', train_cat), ('train_dat_96', train_dat), ('train_info', train_info), ('test_cat', test_cat), ('test_dat_96', test_dat), ('test_info', test_info))
for fname, tensor in files:
print 'Saving to ', fname, '...'
with gzip.open(path+fname+'.pkl.gz','wb') as f:
cPickle.dump(tensor, f)
# Save downscaled version too
w = 48
files = (('test_dat', test_dat),)
for fname, tensor in files:
print 'Generating downscaled version ' + fname + '...'
left = reshape_images(tensor[:,0,:,:], (w,w))
right = reshape_images(tensor[:,1,:,:], (w,w))
result = np.zeros((tensor.shape[0], 2, w, w), dtype=np.uint8)
result[:,0,:,:] = left
result[:,1,:,:] = right
f = gzip.open(path+fname+'_'+str(w)+'.pkl.gz', 'wb')
cPickle.dump(result, f)
f.close()
def load_filetensor(fname):
f = None
try:
if not os.path.exists(fname):
fname = fname+'.gz'
f = gzip.open(fname)
elif fname.endswith('.gz'):
f = gzip.open(fname)
else:
f = open(fname)
d = ft.read(f)
finally:
if f:
f.close()
return d
def convert_orig_to_np():
from pylearn2.datasets.filetensor import read
import gzip
import cPickle
# Load data
path_orig = './data/small_norb/mat/'
prefix_train = path_orig+'smallnorb-5x46789x9x18x6x2x96x96-training-'
train_cat = read(gzip.open(prefix_train+'cat.mat.gz'))
train_dat = read(gzip.open(prefix_train+'dat.mat.gz'))
train_info = read(gzip.open(prefix_train+'info.mat.gz'))
prefix_test = path_orig+'smallnorb-5x01235x9x18x6x2x96x96-testing-'
test_cat = read(gzip.open(prefix_test+'cat.mat.gz'))
test_dat = read(gzip.open(prefix_test+'dat.mat.gz'))
test_info = read(gzip.open(prefix_test+'info.mat.gz'))
# Save originals matrices to file
files = (('train_cat', train_cat), ('train_dat_96', train_dat), ('train_info', train_info), ('test_cat', test_cat), ('test_dat_96', test_dat), ('test_info', test_info))
for fname, tensor in files:
print 'Saving to ', fname, '...'
with gzip.open(path+fname+'.pkl.gz','wb') as f:
cPickle.dump(tensor, f)
# Save downscaled version too
w = 48
files = (('train_dat', train_dat),('test_dat', test_dat))
for fname, tensor in files:
print 'Generating downscaled version ' + fname + '...'
left = reshape_images(tensor[:,0,:,:], (w,w))
right = reshape_images(tensor[:,1,:,:], (w,w))
result = np.zeros((tensor.shape[0], 2, w,w), dtype=np.uint8)
result[:,0,:,:] = left
result[:,1,:,:] = right
f = gzip.open(path+fname+'_'+str(w)+'.pkl.gz', 'wb')
cPickle.dump(result, f)
f.close()
w = 32
files = (('train_dat', train_dat),('test_dat', test_dat))
for fname, tensor in files:
print 'Generating downscaled version ' + fname + '...'
left = reshape_images(tensor[:,0,:,:], (w,w))
right = reshape_images(tensor[:,1,:,:], (w,w))
result = np.zeros((tensor.shape[0], 2, w, w), dtype=np.uint8)
result[:,0,:,:] = left
result[:,1,:,:] = right
f = gzip.open(path+fname+'_'+str(w)+'.pkl.gz', 'wb')
cPickle.dump(result, f)
f.close()
