def train_dataloader(self):
with open('Samples/' + self.dataset + '_train_samples.pickle',
'rb') as handle:
train = pickle.load(handle)
train_dataset = SignedPairsDataset(train, get_index_dicts(train))
return DataLoader(train_dataset,
batch_size=128,
shuffle=True,
num_workers=10,
collate_fn=lambda b: train_dataset.collate(
b,
tcr_encoding=self.tcr_encoding_model,
cat_encoding=self.cat_encoding))
def auc_predict(model, test, train_dicts, peptide=None):
if peptide:
test_dataset = SinglePeptideDataset(test,
train_dicts,
peptide,
force_peptide=False)
else:
test_dataset = SignedPairsDataset(test, train_dicts)
# print(test_dataset.data)
loader = DataLoader(test_dataset,
batch_size=64,
shuffle=False,
num_workers=0,
collate_fn=lambda b: test_dataset.collate(
b,
tcr_encoding=model.tcr_encoding_model,
cat_encoding=model.cat_encoding))
outputs = []
for batch_idx, batch in enumerate(loader):
output = model.validation_step(batch, batch_idx)
if output:
outputs.append(output)
# print(output['y'])
auc = model.validation_end(outputs)['val_auc']
return auc
def predict(dataset, test_file):
model, train_file = get_model(dataset)
train_dicts = get_train_dicts(train_file)
test_samples, dataframe = read_input_file(test_file)
test_dataset = SignedPairsDataset(test_samples, train_dicts)
batch_size = 1000
loader = DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
collate_fn=lambda b: test_dataset.collate(
b,
tcr_encoding=model.tcr_encoding_model,
cat_encoding=model.cat_encoding))
outputs = []
for batch_idx, batch in enumerate(loader):
output = model.validation_step(batch, batch_idx)
if output:
outputs.extend(output['y_hat'].tolist())
dataframe['Score'] = outputs
return dataframe