def get_scores(args, split='test'):
if split == 'train':
input_file = os.path.join(args.data_dir, args.csvtrain)
filescores = os.path.join(
args.data_dir, 'PriorScores/priorscores_answers_train.pckl')
elif split == 'val':
input_file = os.path.join(args.data_dir, args.csvval)
filescores = os.path.join(args.data_dir,
'PriorScores/priorscores_answers_val.pckl')
elif split == 'test':
input_file = os.path.join(args.data_dir, args.csvtest)
filescores = os.path.join(args.data_dir,
'PriorScores/priorscores_answers_test.pckl')
if os.path.exists(filescores):
return
# Load Model
tokenizer = BertTokenizer.from_pretrained(args.bert_model,
do_lower_case=args.do_lower_case)
output_model_file = os.path.join(outdir, WEIGHTS_NAME)
output_config_file = os.path.join(outdir, CONFIG_NAME)
config = BertConfig(output_config_file)
model = BertForMultipleChoice(config, num_choices=4)
model.load_state_dict(torch.load(output_model_file))
model.to(args.device)
n_gpu = torch.cuda.device_count()
logger.info("device: {} n_gpu: {}".format(args.device, n_gpu))
if n_gpu > 1:
model = torch.nn.DataParallel(model)
# Data
eval_examples = read_samples(input_file)
eval_features = convert_to_features(eval_examples, tokenizer,
args.max_seq_length)
all_input_ids = torch.tensor(select_field(eval_features, 'input_ids'),
dtype=torch.long)
all_input_mask = torch.tensor(select_field(eval_features, 'input_mask'),
dtype=torch.long)
all_segment_ids = torch.tensor(select_field(eval_features, 'segment_ids'),
dtype=torch.long)
all_labels = torch.tensor([example.label for example in eval_examples],
dtype=torch.long)
eval_data = TensorDataset(all_input_ids, all_input_mask, all_segment_ids,
all_labels)
eval_sampler = SequentialSampler(eval_data)
eval_dataloader = DataLoader(eval_data,
sampler=eval_sampler,
batch_size=args.eval_batch_size)
# Run prediction
logger.info("***** Compute prior scores *****")
logger.info("Num examples = %d", len(eval_examples))
logger.info("Batch size = %d", args.eval_batch_size)
model.eval()
batch_idx = 0
for _, batch in enumerate(tqdm(eval_dataloader, desc="Iteration")):
input_ids, input_mask, segment_ids, truelabel = batch
input_ids = input_ids.to(args.device)
input_mask = input_mask.to(args.device)
segment_ids = segment_ids.to(args.device)
with torch.no_grad():
logits = model(input_ids, segment_ids, input_mask)
logits = nn.functional.softmax(logits)
logits = logits.detach().cpu().numpy()
if batch_idx == 0:
scores = logits
else:
scores = np.concatenate((scores, logits), axis=0)
batch_idx += 1
if not os.path.exists(os.path.dirname(filescores)):
os.mkdir(os.path.dirname(filescores))
utils.save_obj(scores, filescores)
logger.info('Prior scores for %s saved into %s' % (split, filescores))
def main():
parser = argparse.ArgumentParser()
#drive.mount('/content/gdrive')
swagDir = './data'
cacheDir = './cache/'
saveDir = './save/cache/'
modelDir = './save/cache/pytorch_model.bin'
## Required parameters
parser.add_argument(
"--data_dir",
default=swagDir,
type=str,
#required=True,
help=
"The input data dir. Should contain the .csv files (or other data files) for the task."
)
parser.add_argument(
"--bert_model",
default="bert-base-uncased",
type=str,
#required=True,
help="Bert pre-trained model selected in the list: bert-base-uncased, "
"bert-large-uncased, bert-base-cased, bert-large-cased, bert-base-multilingual-uncased, "
"bert-base-multilingual-cased, bert-base-chinese.")
parser.add_argument(
"--output_dir",
default=saveDir,
type=str,
#required=True,
help="The output directory where the model checkpoints will be written."
)
## Other parameters
parser.add_argument(
"--max_seq_length",
default=100,
type=int,
help=
"The maximum total input sequence length after WordPiece tokenization. \n"
"Sequences longer than this will be truncated, and sequences shorter \n"
"than this will be padded.")
parser.add_argument("--do_train",
default=False,
action='store_true',
help="Whether to run training.")
parser.add_argument("--do_eval",
default=True,
action='store_true',
help="Whether to run eval on the dev set.")
parser.add_argument(
"--do_lower_case",
action='store_true',
help="Set this flag if you are using an uncased model.")
parser.add_argument("--train_batch_size",
default=1,
type=int,
help="Total batch size for training.")
parser.add_argument("--eval_batch_size",
default=1,
type=int,
help="Total batch size for eval.")
parser.add_argument("--learning_rate",
default=5e-5,
type=float,
help="The initial learning rate for Adam.")
parser.add_argument("--num_train_epochs",
default=3.0,
type=float,
help="Total number of training epochs to perform.")
parser.add_argument(
"--warmup_proportion",
default=0.1,
type=float,
help=
"Proportion of training to perform linear learning rate warmup for. "
"E.g., 0.1 = 10%% of training.")
parser.add_argument("--no_cuda",
action='store_true',
help="Whether not to use CUDA when available")
parser.add_argument("--local_rank",
type=int,
default=-1,
help="local_rank for distributed training on gpus")
parser.add_argument('--seed',
type=int,
default=42,
help="random seed for initialization")
parser.add_argument(
'--gradient_accumulation_steps',
type=int,
default=1,
help=
"Number of updates steps to accumulate before performing a backward/update pass."
)
parser.add_argument(
'--fp16',
action='store_true',
help="Whether to use 16-bit float precision instead of 32-bit")
parser.add_argument(
'--loss_scale',
type=float,
default=0,
help=
"Loss scaling to improve fp16 numeric stability. Only used when fp16 set to True.\n"
"0 (default value): dynamic loss scaling.\n"
"Positive power of 2: static loss scaling value.\n")
args = parser.parse_args()
device = torch.device("cuda")
n_gpu = torch.cuda.device_count()
"""
if args.local_rank == -1 or args.no_cuda:
device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu")
n_gpu = torch.cuda.device_count()
else:
torch.cuda.set_device(args.local_rank)
device = torch.device("cuda", args.local_rank)
n_gpu = 1
# Initializes the distributed backend which will take care of sychronizing nodes/GPUs
torch.distributed.init_process_group(backend='nccl')
"""
logger.info(
"device: {} n_gpu: {}, distributed training: {}, 16-bits training: {}".
format(device, n_gpu, bool(args.local_rank != -1), args.fp16))
if args.gradient_accumulation_steps < 1:
raise ValueError(
"Invalid gradient_accumulation_steps parameter: {}, should be >= 1"
.format(args.gradient_accumulation_steps))
args.train_batch_size = args.train_batch_size // args.gradient_accumulation_steps
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if n_gpu > 0:
torch.cuda.manual_seed_all(args.seed)
if not args.do_train and not args.do_eval:
raise ValueError(
"At least one of `do_train` or `do_eval` must be True.")
"""
if os.path.exists(args.output_dir) and os.listdir(args.output_dir):
raise ValueError("Output directory ({}) already exists and is not empty.".format(args.output_dir))
if not os.path.exists(args.output_dir):
os.makedirs(args.output_dir)
"""
tokenizer = BertTokenizer.from_pretrained(args.bert_model,
do_lower_case=args.do_lower_case)
train_examples = None
num_train_optimization_steps = None
# Prepare model
model = BertForMultipleChoice.from_pretrained(
args.bert_model,
cache_dir=os.path.join(str(PYTORCH_PRETRAINED_BERT_CACHE),
'distributed_{}'.format(args.local_rank)),
num_choices=4)
model.to(device)
if args.local_rank != -1:
try:
from apex.parallel import DistributedDataParallel as DDP
except ImportError:
raise ImportError(
"Please install apex from https://www.github.com/nvidia/apex to use distributed and fp16 training."
)
model = DDP(model)
elif n_gpu > 1:
model = torch.nn.DataParallel(model)
# Prepare optimizer
param_optimizer = list(model.named_parameters())
# hack to remove pooler, which is not used
# thus it produce None grad that break apex
param_optimizer = [n for n in param_optimizer if 'pooler' not in n[0]]
no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']
optimizer_grouped_parameters = [{
'params':
[p for n, p in param_optimizer if not any(nd in n for nd in no_decay)],
'weight_decay':
0.01
}, {
'params':
[p for n, p in param_optimizer if any(nd in n for nd in no_decay)],
'weight_decay':
0.0
}]
"""
if args.do_train:
# Save a trained model and the associated configuration
model_to_save = model.module if hasattr(model, 'module') else model # Only save the model it-self
output_model_file = os.path.join(args.output_dir, WEIGHTS_NAME)
torch.save(model_to_save.state_dict(), output_model_file)
output_config_file = os.path.join(args.output_dir, CONFIG_NAME)
with open(output_config_file, 'w') as f:
f.write(model_to_save.config.to_json_string())
# Load a trained model and config that you have fine-tuned
config = BertConfig(output_config_file)
model = BertForMultipleChoice(config, num_choices=4)
model.load_state_dict(torch.load(output_model_file))
else:
model = BertForMultipleChoice.from_pretrained(args.bert_model, num_choices=4)
model.to(device)
"""
#output_model_file = os.path.join(args.output_dir, WEIGHTS_NAME)
ckpt1 = "./mrpcckpt/pytorch_model.bin"
ckpt2 = "./mnlickpt/pytorch_model.bin"
ckpt3 = "./sstckpt/pytorch_model.bin"
output_config_file = os.path.join(args.output_dir, CONFIG_NAME)
config = BertConfig(output_config_file)
model1 = BertForMultipleChoice(config, num_choices=4).to(device)
model1.load_state_dict(torch.load(ckpt1))
model2 = BertForMultipleChoice(config, num_choices=4).to(device)
model2.load_state_dict(torch.load(ckpt2))
model3 = BertForMultipleChoice(config, num_choices=4).to(device)
model3.load_state_dict(torch.load(ckpt3))
#print(device)
if args.do_eval and (args.local_rank == -1
or torch.distributed.get_rank() == 0):
eval_examples = read_swag_examples(os.path.join(
args.data_dir, 'val.csv'),
is_training=True)
eval_features = convert_examples_to_features(eval_examples, tokenizer,
args.max_seq_length, True)
logger.info("***** Running evaluation *****")
logger.info(" Num examples = %d", len(eval_examples))
logger.info(" Batch size = %d", args.eval_batch_size)
all_input_ids = torch.tensor(select_field(eval_features, 'input_ids'),
dtype=torch.long)
all_input_mask = torch.tensor(select_field(eval_features,
'input_mask'),
dtype=torch.long)
all_segment_ids = torch.tensor(select_field(eval_features,
'segment_ids'),
dtype=torch.long)
all_label = torch.tensor([f.label for f in eval_features],
dtype=torch.long)
eval_data = TensorDataset(all_input_ids, all_input_mask,
all_segment_ids, all_label)
# Run prediction for full data
eval_sampler = SequentialSampler(eval_data)
eval_dataloader = DataLoader(eval_data,
sampler=eval_sampler,
batch_size=args.eval_batch_size)
model1.eval()
model2.eval()
model3.eval()
eval_loss, eval_accuracy = 0, 0
nb_eval_steps, nb_eval_examples = 0, 0
evalcount = 0
for input_ids, input_mask, segment_ids, label_ids in tqdm(
eval_dataloader, desc="Evaluating"):
input_ids = input_ids.to(device)
input_mask = input_mask.to(device)
segment_ids = segment_ids.to(device)
label_ids = label_ids.to(device)
with torch.no_grad():
tmp_eval_loss = model(input_ids, segment_ids, input_mask,
label_ids)
logits1 = model1(input_ids, segment_ids, input_mask)
logits2 = model2(input_ids, segment_ids, input_mask)
logits3 = model3(input_ids, segment_ids, input_mask)
logits = logits1 + logits2 + logits3
logits = logits.detach().cpu().numpy()
label_ids = label_ids.to('cpu').numpy()
tmp_eval_accuracy = accuracy(logits, label_ids)
"""
if tmp_eval_accuracy ==0:
print("Wrong prediction example No. ", evalcount, "with prediction", np.argmax(logits, axis=1), "and groundtruth", label_ids)
print("Context sentence: ")
print(eval_examples[evalcount].context_sentence)
print("Start ending:")
print(eval_examples[evalcount].start_ending)
print("Candidate answer 0:")
print(eval_examples[evalcount].endings[0])
print("Candidate answer 1:")
print(eval_examples[evalcount].endings[1])
print("Candidate answer 2:")
print(eval_examples[evalcount].endings[2])
print("Candidate answer 3:")
print(eval_examples[evalcount].endings[3])
print(eval_examples[evalcount].label)
"""
if tmp_eval_accuracy == 1:
print(evalcount)
eval_loss += tmp_eval_loss.mean().item()
eval_accuracy += tmp_eval_accuracy
nb_eval_examples += input_ids.size(0)
nb_eval_steps += 1
evalcount += 1
eval_loss = eval_loss / nb_eval_steps
eval_accuracy = eval_accuracy / nb_eval_examples
result = {
'eval_loss': eval_loss,
'eval_accuracy': eval_accuracy,
}
output_eval_file = os.path.join(args.output_dir, "eval_results.txt")
with open(output_eval_file, "w") as writer:
logger.info("***** Eval results *****")
for key in sorted(result.keys()):
logger.info(" %s = %s", key, str(result[key]))
writer.write("%s = %s\n" % (key, str(result[key])))
