def evaluate(checkpoint: str, data: str = None, batch_size: int = None):
seed_everything(SEED)
model = Code2Seq.load_from_checkpoint(checkpoint_path=checkpoint)
batch_size = batch_size or model.hyperparams.test_batch_size
data = data or model.hyperparams.test_data_path
gpu = 1 if torch.cuda.is_available() else None
data_loader, n_samples = create_dataloader(
data, model.hyperparams.max_context, False, False, batch_size, cpu_count(),
)
print(f"approximate number of steps for test is {ceil(n_samples / batch_size)}")
trainer = Trainer(gpus=gpu)
trainer.test(model, test_dataloaders=data_loader)
def evaluate(checkpoint: str, data: str = None):
seed_everything(SEED)
model = Code2Seq.load_from_checkpoint(checkpoint_path=checkpoint)
gpu = 1 if torch.cuda.is_available() else None
trainer = Trainer(gpus=gpu)
if data is not None:
data_loader, n_samples = create_dataloader(
join(DATA_FOLDER, data), model.config.max_context, False, False,
model.config.test_batch_size, cpu_count())
print(
f"approximate number of steps for test is {ceil(n_samples / model.config.test_batch_size)}"
)
trainer.test(model, test_dataloaders=data_loader)
else:
trainer.test(model)
def load_code2seq(
checkpoint_path: str, config: DictConfig,
vocabulary: Vocabulary) -> Tuple[Code2Seq, PathContextDataModule]:
model = Code2Seq.load_from_checkpoint(checkpoint_path=checkpoint_path)
data_module = PathContextDataModule(config, vocabulary)
return model, data_module
print('Done dumping reduced data set')
return out_path
if __name__=="__main__":
opt = parse_args()
print(opt)
print('data path: ', opt.data_path)
data_split = opt.data_path.split('/')[-2]
print('data_split', data_split)
# replace_tokens = ["@R_%d@"%x for x in range(0,opt.num_replacements+1)]
replace_tokens = ["@R_%d@"%x for x in range(1000)]
model = Code2Seq.load_from_checkpoint(checkpoint_path=opt.expt_dir)
data_loader, n_samples = create_dataloader(
opt.data_path, model.hyperparams.max_context, False, False, opt.batch_size, 1,
)
vocab = pickle.load(open(opt.vocab, 'rb'))
token_to_id = vocab['token_to_id']
id_to_token = {token_to_id[t]:t for t in token_to_id}
print('length: ', len(id_to_token))
label_to_id = vocab['label_to_id']
id_to_label = {label_to_id[t]:t for t in label_to_id}
# if data_split == 'test' and opt.exact_matches:
args = parser.parse_args()
return args
def create_datafile(data_path, exact_matches, split):
new_data_path = os.path.join(data_path, 'small.{}.c2s'.format(split))
lines = open(os.path.join(data_path, 'data.{}.c2s'.format(split)), 'r')
new_file = open(new_data_path, 'w')
for line in lines:
if line.split()[0] in exact_matches:
new_file.write(line)
print("Saved exact matches.")
if __name__ == '__main__':
args = parse_args()
model = Code2Seq.load_from_checkpoint(checkpoint_path=args.checkpoint)
data_loader, n_samples = create_dataloader(
os.path.join(args.orig_data_path, args.split),
model.hyperparams.max_context, False, False, args.batch_size, 1)
vocab = pickle.load(open(args.vocab_path, 'rb'))
label_to_id = vocab['label_to_id']
id_to_label = {label_to_id[l]: l for l in label_to_id}
li_exact_matches = get_exact_matches(data_loader, n_samples, model,
id_to_label)
print(li_exact_matches)
create_datafile(args.data_path, li_exact_matches, args.split)