def from_conf(cls,
conf,
_max_expe=2e6,
private_keywords=[],
expdesign=None):
gt = cls(private_keywords=private_keywords)
_spec = conf.pop('_spec', None)
if not _spec:
if not expdesign:
expdesign = ExpDesign
conf['_name_expe'] = '_default_expe'
conf['_expe_hash'] = hash_objects(
dict((k, v) for k, v in conf.items()
if k not in private_keywords))
gt._tensors.append(ExpTensor.from_expe(conf))
gt._ds_.append(expdesign)
return gt
exp = []
size_expe = len(_spec)
consume_expe = 0
while consume_expe < size_expe:
o = _spec[consume_expe]
if isinstance(o, tuple):
# _type => expdesign
name, o, _type = o
if isinstance(o, ExpGroup):
size_expe += len(o) - 1
_spec = _spec[:consume_expe] + o + _spec[consume_expe + 1:]
elif isinstance(o, list):
exp.append(o)
gt._ds_.append(_type)
consume_expe += 1
else:
o['_name_expe'] = name
o['_expe_hash'] = hash_objects(
dict((k, v) for k, v in o.items()
if k not in private_keywords))
exp.append(o)
gt._ds_.append(_type)
consume_expe += 1
if size_expe > _max_expe:
lgg.warning(
'Number of experiences exceeds the hard limit of %d (please review ExpTensor).'
% _max_expe)
gt._tensors.extend([ExpTensor.from_expe(conf, spec) for spec in exp])
return gt
def fit(self):
voca = Vocabulary(exclude_stopwords=True)
writer = self.get_writer(reset=self.expe.reset, online=True)
setattr(self, 'writer', writer)
for _it, path in enumerate(self.doc_yielder(self.expe.path)):
fullpath = path
shortpath = '/' + fullpath[len(os.path.expanduser(self.expe.path)):].rstrip('/').lstrip('/')
is_known = False
is_duplicated = False
if self.getfirst(shortpath, 'shortpath'):
# don't update document
# could compute a diff here...
is_known = True # assume already indexed
else:
text = extract_pdf(fullpath)
text = voca.remove_stopwords(text)
#bow = voca.doc2bow(text)
if text in (None, ''):
# do nothing
continue
doc = dict(shortpath=shortpath, fullpath=fullpath)
doc['content'] = text
doc['hash'] = hash_objects(text)
first_m = self.getfirst(doc['hash'], 'hash')
if first_m:
#if not 'content' in first_m:
# writer.delete_by_term('hash', doc['hash'])
# continue
# don't update document
self.log.warning("Duplicate file detected: %s renaming to %s" % (first_m['shortpath'], shortpath))
first_m['shortpath'] = shortpath
writer.update_document(**first_m)
is_duplicated = True
else:
if self.expe.extract_structure:
# structured content
structured = self.extract_structured_kw(doc)
doc.update(structured)
if not (is_known or is_duplicated):
print("indexing `%s'" % (path))
try:
writer.add_document(**doc)
except Exception as e:
print('indexing doc %s failed!' % fullpath)
return
def _make_hash(self, skip_check=False):
_hash = []
n_duplicate = 0
for _id, _d in enumerate(self._lod):
d = _d.copy()
[ d.pop(k) for k in self._private_keywords if k in d and k != '_repeat']
o = hash_objects(d)
if o in _hash:
n_duplicate += 1
_hash.append(o)
if n_duplicate > 0 and not skip_check:
lgg.warning('Duplicate experience: %d' % (n_duplicate))
ask_sure_exit('Continue [y/n]?')
self._hash = _hash
def gen_data_matlab(self):
from scipy.io import savemat
from pymake.util.utils import hash_objects
expe = self.expe
expe['driver'] = 'gt'
training_ratio = 100
testset_ratio = 20
validset_ratio = 10
corpus_name = expe.corpus
training_ratio = str(int(expe.get('training_ratio', training_ratio)))
testset_ratio = str(int(expe.get('testset_ratio', testset_ratio)))
validset_ratio = str(int(expe.get('validset_ratio', validset_ratio)))
repeat = expe.get('_repeat', '')
expe['training_ratio'] = training_ratio
expe['testset_ratio'] = testset_ratio
expe['validset_ratio'] = validset_ratio
# Debug how validset is computed
#expe['testset_ratio'] -= 0.1/1.1
frontend = self.load_frontend()
is_symmetric = frontend.is_symmetric()
### sparse matrix with test indexes at 1
Ytest = frontend.data_test
### Adjacency matrix
g = frontend.data
g.clear_filters()
y = frontend.adj()
# set the weight
for i, j, w in frontend.get_edges():
y[i, j] = w
if is_symmetric:
y[j, i] = w
# state
seed = []
for c in list(corpus_name):
seed.append(str(ord(c)))
seed.append(repeat)
seed = ''.join([chr(int(i)) for i in list(''.join(seed))])
seed = int((hash_objects(seed)), 32) % 2**32
out = os.path.join(self.get_data_path(), 'mat') + '/'
if repeat:
out += repeat + '/'
os.makedirs(out, exist_ok=True)
fnout = out + corpus_name + '_' + '-'.join(
[training_ratio, testset_ratio, validset_ratio]) + '.mat'
print('saving: %s' % fnout)
savemat(
fnout, {
'Y': y.astype(float),
'Ytest': Ytest.astype(float),
'is_symmetric': is_symmetric,
'state': seed
})
