def arg_l_arg_r_pairs_vector(args_file, file_contains_context=False, has_header=True):
"""
"""
logging.info("creating arg pairs class vector '{}'".format(args_file))
temp = []
if file_contains_context:
ctx_argl_argr_entail = tio.read_args_w_ctx(args_file, has_header=has_header)
else:
argl_argr_entail = tio.read_args_wo_ctx(args_file, has_header=has_header)
def append_empty_context(tuples):
for l,r,e in tuples:
yield '', l, r, e
ctx_argl_argr_entail = append_empty_context(argl_argr_entail)
d_triples = td.TripleDict() # rows
duplicates = 0
contradicting_duplicates = 0
for ctx, arg_l, arg_r, entailing in ctx_argl_argr_entail:
i = d_triples.add((ctx, arg_l, arg_r))
if i < len(temp):
label = 1 if entailing.strip().lower() == 'true' else 0
print("omitting duplicate example: '{} {} {} {}' ".format(ctx, arg_l, arg_r, entailing) ,file=sys.stderr)
duplicates += 1
if temp[i] != label:
print("duplicate example has different label: '{}' vs. '{}'".format(temp[i], label) ,file=sys.stderr)
contradicting_duplicates += 1
else:
temp.append(1 if entailing.strip().lower() == 'true' else 0)
vec = np.array(temp, dtype=np.float64)
logging.info("finished creating arg pairs class vector '{}'".format(args_file))
logging.info("found {} duplicate examples with {} having contradicting labels.".format(duplicates, contradicting_duplicates))
return vec, d_triples
def get_train_test_indexes_presplit(d_triples):
"""
"""
t1 = tio.read_args_w_ctx('../data/updates/4/args_v2_am.tsv', has_header=False)
train_ids = []
for ctx, arg_l, arg_r, __ in t1:
id_ = d_triples.get_triple_id((ctx, arg_l, arg_r))
train_ids.append(id_)
t2 = tio.read_args_w_ctx('../data/updates/4/args_v2_nz.tsv', has_header=False)
test_ids = []
for ctx, arg_l, arg_r, __ in t2:
id_ = d_triples.get_triple_id((ctx, arg_l, arg_r))
test_ids.append(id_)
return train_ids, test_ids
def get_train_test_indexes_presplit(d_triples):
"""
"""
t1 = tio.read_args_w_ctx('../data/updates/4/args_v2_am.tsv',
has_header=False)
train_ids = []
for ctx, arg_l, arg_r, __ in t1:
id_ = d_triples.get_triple_id((ctx, arg_l, arg_r))
train_ids.append(id_)
t2 = tio.read_args_w_ctx('../data/updates/4/args_v2_nz.tsv',
has_header=False)
test_ids = []
for ctx, arg_l, arg_r, __ in t2:
id_ = d_triples.get_triple_id((ctx, arg_l, arg_r))
test_ids.append(id_)
return train_ids, test_ids
def arg_l_arg_r_pairs_vector(args_file,
file_contains_context=False,
has_header=True):
"""
"""
logging.info("creating arg pairs class vector '{}'".format(args_file))
temp = []
if file_contains_context:
ctx_argl_argr_entail = tio.read_args_w_ctx(args_file,
has_header=has_header)
else:
argl_argr_entail = tio.read_args_wo_ctx(args_file,
has_header=has_header)
def append_empty_context(tuples):
for l, r, e in tuples:
yield '', l, r, e
ctx_argl_argr_entail = append_empty_context(argl_argr_entail)
d_triples = td.TripleDict() # rows
duplicates = 0
contradicting_duplicates = 0
for ctx, arg_l, arg_r, entailing in ctx_argl_argr_entail:
i = d_triples.add((ctx, arg_l, arg_r))
if i < len(temp):
label = 1 if entailing.strip().lower() == 'true' else 0
print("omitting duplicate example: '{} {} {} {}' ".format(
ctx, arg_l, arg_r, entailing),
file=sys.stderr)
duplicates += 1
if temp[i] != label:
print("duplicate example has different label: '{}' vs. '{}'".
format(temp[i], label),
file=sys.stderr)
contradicting_duplicates += 1
else:
temp.append(1 if entailing.strip().lower() == 'true' else 0)
vec = np.array(temp, dtype=np.float64)
logging.info(
"finished creating arg pairs class vector '{}'".format(args_file))
logging.info(
"found {} duplicate examples with {} having contradicting labels.".
format(duplicates, contradicting_duplicates))
return vec, d_triples
