def test_num_entries_empty(self):
path_lmdb_empty = os.path.join(self.dir_tmp, 'empty_lmdb')
db = lmdb.open(path_lmdb_empty, map_size=int(1e12))
with db.begin(write=True) as _:
_
db.close()
assert_equal(0, r.num_entries(path_lmdb_empty))
def est_min_num_fwd_passes(fpath_net, mode_str):
"""
if multiple source for same mode, base num_passes on last
fpath_net -- path to network definition
mode_str -- train or test?
return
minimum no. of forward passes to cover training set
"""
from proto.proto_utils import Parser
np = Parser().from_net_params_file(fpath_net)
num_passes = 0
for l in np.layer:
if 'data' in l.type.lower() and mode_str.lower() in l.data_param.source.lower():
num_entries = read_lmdb.num_entries(l.data_param.source)
num_passes = int(num_entries / l.data_param.batch_size)
if num_entries % l.data_param.batch_size != 0:
print("WARNING: db size not a multiple of batch size. Adding another fwd. pass.")
num_passes += 1
print("%d fwd. passes with batch size %d" % (num_passes, l.data_param.batch_size))
return num_passes
def test_num_entries_non_num(self):
assert_equal(2, r.num_entries(self.path_lmdb_non_num))
assert_equal(2, r.num_entries(self.path_lmdb_non_num, is_num_ord_dense=False))
assert_raises(ValueError, r.num_entries, self.path_lmdb_non_num, is_num_ord_dense=True)
def test_num_entries_rand_ord(self):
assert_equal(2, r.num_entries(self.path_lmdb_rand_ord))
assert_equal(2, r.num_entries(self.path_lmdb_rand_ord, is_num_ord_dense=False))
assert_not_equal(2, r.num_entries(self.path_lmdb_rand_ord, is_num_ord_dense=True))
def test_num_entries(self):
assert_equal(2, r.num_entries(self.path_lmdb))
assert_equal(2, r.num_entries(self.path_lmdb, is_num_ord_dense=True))
assert_equal(2, r.num_entries(self.path_lmdb, is_num_ord_dense=False))
out[m] = infer_to_lmdb(caffe.Net(fpath_net, fpath_weights, m),
keys,
num_passes,
dst_prefix + '%s_' + ['train', 'test'][m] + '_lmdb')
return out
if __name__ == '__main__':
base_path = '/mnt/scratch/pierre/caffe_sandbox_tryouts/'
def path_to(path):
return base_path + path
fpath_net = path_to('learning_curve/prototxt/net0_test.prototxt')
fpath_weights = path_to('learning_curve/snapshots/net0_snapshot_iter_10000.caffemodel')
fpath_db = path_to('inference/mnist_%s_train_lmdb')
net = caffe.Net(fpath_net, fpath_weights, caffe.TRAIN)
keys = ['fc2', 'fc1']
x = infer_to_lmdb_cur(net, keys, 2, fpath_db)
import os
print 'Do lmdbs exist?'
print [os.path.isdir(fpath_db % (k,)) for k in keys]
print 'Number of entries in lmdbs:'
print [read_lmdb.num_entries(fpath_db % (k,)) for k in keys]
# Here you need to compute accuracy, roc, precision recall, confusion matrix
pass
fpath_net = expanduser('~/models/dark/mnist/t0/lenet_train_test.prototxt')
fpath_weights = expanduser('~/models/dark/mnist/t0/lenet_iter_10000.caffemodel')
# x = response_to_lmdb(fpath_net, fpath_weights,
# ['ip2', 'ip1'],
# expanduser('~/models/dark/mnist/t0/mnistX_'))
net = caffe.Net(fpath_net, fpath_weights, caffe.TRAIN)
keys = ['ip2', 'ip1']
x = infer_to_lmdb_cur(net, keys, 2,
expanduser('~/models/dark/mnist/t0/Xmnist_%s_train_lmdb'))
print x
import os
print [os.path.isdir(expanduser('~/models/dark/mnist/t0/Xmnist_%s_train_lmdb') % (k,)) for k in keys]
print [read_lmdb.num_entries(expanduser('~/models/dark/mnist/t0/Xmnist_%s_train_lmdb') % (k,)) for k in keys]
#print [read_lmdb.read_values(expanduser('~/models/dark/mnist/t0/Xmnist_%s_train_lmdb') % (k,)) for k in keys]
# with h5py.File(fpath, "w") as f:
#
# f['a'] = 0
#
#
# f['b'] = [1, 2]
# f['c'] = np.arange(3)
# f['d'] = [np.array([[1,2],[4,5]], dtype=float), np.array([[1,2],[4, 5]], dtype=float)+10]
#infer_to_h5(net, 1, ['accuracy'], fpath)
pass
