def __getitem__(self, idx):
batch_index = np.arange(idx * self.batch_size,
min(self.size, (idx + 1) * self.batch_size))
df = self.df.iloc[batch_index]
data_seq = np.zeros((len(df), MAXLEN, 21), dtype=np.float32)
data_gos = np.zeros((len(df), len(self.gos_dict)), dtype=np.float32)
data_exp = np.zeros((len(df), 53), dtype=np.float32)
labels = np.zeros((len(df), len(self.terms_dict)), dtype=np.int32)
for i, row in enumerate(df.itertuples()):
data_seq[i, :] = to_onehot(row.sequences)
data_exp[i, :] = row.expressions
for item in row.deepgo_annotations:
t_id, score = item.split('|')
if t_id in self.gos_dict:
data_gos[i, self.gos_dict[t_id]] = float(score)
for t_id in row.iea_annotations:
if t_id in self.gos_dict:
data_gos[i, self.gos_dict[t_id]] = 1
for t_id in row.go_annotations:
if t_id in self.gos_dict:
data_gos[i, self.gos_dict[t_id]] = 1
for t_id in row.hp_annotations:
if t_id in self.terms_dict:
labels[i, self.terms_dict[t_id]] = 1
data = np.concatenate([data_gos, data_exp], axis=1)
return (data, labels)
def __getitem__(self, idx):
batch_index = np.arange(idx * self.batch_size,
min(self.size, (idx + 1) * self.batch_size))
df = self.df.iloc[batch_index]
data_onehot = np.zeros((len(df), MAXLEN, 21), dtype=np.float32)
labels = np.zeros((len(df), self.nb_classes), dtype=np.int32)
for i, row in enumerate(df.itertuples()):
seq = row.sequences
onehot = to_onehot(seq)
data_onehot[i, :, :] = onehot
for t_id in row.prop_annotations:
if t_id in self.terms_dict:
labels[i, self.terms_dict[t_id]] = 1
self.start += self.batch_size
print(data_onehot, labels)
return (data_onehot, labels)
def next(self):
if self.start < self.size:
batch_index = np.arange(
self.start, min(self.size, self.start + self.batch_size))
df = self.df.iloc[batch_index]
data_onehot = np.zeros((len(df), MAXLEN, 21), dtype=np.int32)
labels = np.zeros((len(df), self.nb_classes), dtype=np.int32)
for i, row in enumerate(df.itertuples()):
seq = row.sequences
onehot = to_onehot(seq)
data_onehot[i, :, :] = onehot
for t_id in row.prop_annotations:
if t_id in self.terms_dict:
labels[i, self.terms_dict[t_id]] = 1
self.start += self.batch_size
return (data_onehot, labels)
else:
self.reset()
return self.next()
def get_data(sequences):
pred_seqs = []
ids = []
for i, seq in enumerate(sequences):
if len(seq) > MAXLEN:
st = 0
while st < len(seq):
pred_seqs.append(seq[st: st + MAXLEN])
ids.append(i)
st += MAXLEN - 128
else:
pred_seqs.append(seq)
ids.append(i)
n = len(pred_seqs)
data = np.zeros((n, MAXLEN, 21), dtype=np.float32)
for i in range(n):
seq = pred_seqs[i]
data[i, :, :] = to_onehot(seq)
return ids, data
