def __getitem__(self, i):
if not self.random_pairing:
if self.rnn:
instance = preprocess_seq_for_rnn(self.sequences[i], self.num_alphabets, self.cfg, augment=False)
return instance
else:
sys.exit("PLUS-TFM Pfam pre-training without random pairing is not supported")
else:
trial = 0
while(1):
sequence0 = self.sequences[i + trial // 10]
is_samefamily = random.random() > 0.5
if is_samefamily: family = self.structs[i + trial // 10]
else:
while(1):
family = random.randint(0, len(self.structs_idx) - 1)
if family != self.structs[i + trial // 10] or self.sanity_check: break
sequence1 = self.sequences[random.choice(self.structs_idx[family])]
is_samefamily = torch.tensor([is_samefamily])
if self.max_len is None or len(sequence0) + len(sequence1) + 3 <= self.max_len: break
else: trial += 1
if self.rnn:
instance0 = preprocess_seq_for_rnn(sequence0, self.num_alphabets, self.cfg)
instance1 = preprocess_seq_for_rnn(sequence1, self.num_alphabets, self.cfg)
return instance0, instance1, is_samefamily
else:
instance = preprocess_seq_for_tfm(sequence0, sequence1, self.num_alphabets, self.cfg, self.max_len)
return (*instance, is_samefamily)
def __getitem__(self, i):
sequence0, sequence1 = self.sequences0[i], self.sequences1[i]
is_samefamily = self.labels[i]
if self.rnn:
instance0 = preprocess_seq_for_rnn(sequence0, self.num_alphabets, self.cfg, self.augment)
instance1 = preprocess_seq_for_rnn(sequence1, self.num_alphabets, self.cfg, self.augment)
return instance0, instance1, is_samefamily
else:
instance = preprocess_seq_for_tfm(sequence0, sequence1, self.num_alphabets, self.cfg, self.max_len, self.augment)
return (*instance, is_samefamily)
def __getitem__(self, i):
if self.rnn:
instance = preprocess_seq_for_rnn(self.sequences[i], self.num_alphabets, self.cfg, self.augment)
return instance, self.labels[i]
else:
instance_seq = preprocess_seq_for_tfm(self.sequences[i], None, self.num_alphabets, self.cfg, self.max_len, self.augment)
if self.valids is None:
return (*instance_seq, self.labels[i])
else:
instance_label = preprocess_label_for_tfm(self.labels[i], self.valids[i], self.max_len)
return (*instance_seq, *instance_label)
def __getitem__(self, i):
sequence0, sequence1 = self.sequences0[i], self.sequences1[i]
similarity_level = self.labels[i]
if self.rnn:
instance0 = preprocess_seq_for_rnn(sequence0, self.num_alphabets, self.cfg, self.augment)
instance1 = preprocess_seq_for_rnn(sequence1, self.num_alphabets, self.cfg, self.augment)
if self.cmaps0 is not None: return instance0, instance1, similarity_level, self.cmaps0[i], self.cmaps1[i]
else: return instance0, instance1, similarity_level
else:
instance = preprocess_seq_for_tfm(sequence0, sequence1, self.num_alphabets, self.cfg, self.max_len, self.augment)
if self.cmaps0 is None: return (*instance, similarity_level)
else: return (*instance, similarity_level, self.cmaps0[i], self.cmaps1[i])
def __getitem__(self, k):
n = len(self.sequences)
i, j = k // n, k % n
sequence0, sequence1 = self.sequences[i], self.sequences[j]
similarity_level = self.labels[i, j]
if self.rnn:
instance0 = preprocess_seq_for_rnn(sequence0, self.num_alphabets, self.cfg)
instance1 = preprocess_seq_for_rnn(sequence1, self.num_alphabets, self.cfg)
if self.cmaps is not None: return instance0, instance1, similarity_level, self.cmaps[i], self.cmaps[j]
else: return instance0, instance1, similarity_level
else:
instance = preprocess_seq_for_tfm(sequence0, sequence1, self.num_alphabets, self.cfg, self.max_len)
if self.cmaps is not None: return (*instance, similarity_level, self.cmaps[i], self.cmaps[j])
else: return (*instance, similarity_level)
def __getitem__(self, i):
if self.rnn: instance = preprocess_seq_for_rnn(self.sequences[i], self.num_alphabets, self.cfg, augment=False)
else: instance = preprocess_seq_for_tfm(self.sequences[i], None, self.num_alphabets, self.cfg, self.max_len, augment=False)
return instance