def log_to_csv_with_auc_accuracy(y_true, y_pred, y_score, csv_log_file_path, identity_info="dataset"):
labels = [0, 1]
result = precision_recall_fscore_support(y_true, y_pred)
row = []
row.append(identity_info)
# neg
row.append('label ' + str(labels[0]) + ":")
row.append(result[0][0])
row.append(result[1][0])
row.append(result[2][0])
row.append(result[3][0])
row.append(' ')
# pos
row.append('label ' + str(labels[1]) + ":")
row.append(result[0][1])
row.append(result[1][1])
row.append(result[2][1])
row.append(result[3][1])
row.append(' ')
# auc and accuracy
y_pos_score = transform.map_func(y_score, lambda p : p[1])
auc = metrics.roc_auc_score(y_true, y_pos_score)
row.append(auc)
accuracy = metrics.accuracy_score(y_true, y_pred)
row.append(accuracy)
csv_handler.append_row(csv_log_file_path, row)
def log_to_csv_multi_f1(y_true, y_pred, csv_log_file_path, identity_info="dataset"):
result = precision_recall_fscore_support(y_true, y_pred)
row = []
row.append(identity_info)
multi_f1 = list(result[2])
row += multi_f1
csv_handler.append_row(csv_log_file_path, row)
def log_to_csv(y_true, y_pred, csv_log_file_path, identity_info="dataset"):
labels = [0, 1]
result = precision_recall_fscore_support(y_true, y_pred)
row = []
row.append(identity_info)
# neg
row.append('label ' + str(labels[0]) + ":")
row.append(result[0][0])
row.append(result[1][0])
row.append(result[2][0])
row.append(result[3][0])
row.append(' ')
# pos
row.append('label ' + str(labels[1]) + ":")
row.append(result[0][1])
row.append(result[1][1])
row.append(result[2][1])
row.append(result[3][1])
csv_handler.append_row(csv_log_file_path, row)
#('clf', LogisticRegression(class_weight='balanced', random_state=seed, solver='liblinear')),
('clf', LogisticRegression(random_state=seed, solver='liblinear')),
])
text_clf.fit(X_train, y_train)
train_finish_time = time.time()
train_duration = train_finish_time - start_time
print("train time is " + str(train_finish_time - start_time))
print("predicting...")
predicted = text_clf.predict(X_dev)
predicted_proba = text_clf.predict_proba(X_dev)
assert (len(predicted_proba) == len(X_dev))
assert (len(X_dev) == len(y_dev))
print("logging...")
csv_handler.append_row(output_path,
['score_0', 'score_1', 'predict', 'text', 'ground'])
result = []
for i in range(len(predicted_proba)):
score_0 = predicted_proba[i][0]
score_1 = predicted_proba[i][1]
predict = predicted[i]
text = X_dev[i]
ground = y_dev[i]
result.append([score_0, score_1, predict, text, ground])
csv_handler.csv_writelines(output_path, result)
def main():
start_time = time.time()
parser = argparse.ArgumentParser()
## Required parameters
parser.add_argument("--data_dir",
default=None,
type=str,
required=True,
help="The input data dir. Should contain the .tsv files (or other data files) for the task.")
parser.add_argument("--bert_model", default=None, 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("--task_name",
default=None,
type=str,
required=True,
help="The name of the task to train.")
parser.add_argument("--output_dir",
default=None,
type=str,
required=True,
help="The output directory where the model predictions and checkpoints will be written.")
parser.add_argument("--log_file_path",
default=None,
type=str,
required=True,
help="Define log file path.")
## Other parameters
parser.add_argument("--cache_dir",
default="",
type=str,
help="Where do you want to store the pre-trained models downloaded from s3")
parser.add_argument("--max_seq_length",
default=128,
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",
action='store_true',
help="Whether to run training.")
parser.add_argument("--do_eval",
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=32,
type=int,
help="Total batch size for training.")
parser.add_argument("--eval_batch_size",
default=8,
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")
parser.add_argument('--server_ip', type=str, default='', help="Can be used for distant debugging.")
parser.add_argument('--server_port', type=str, default='', help="Can be used for distant debugging.")
args = parser.parse_args()
if args.server_ip and args.server_port:
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
import ptvsd
print("Waiting for debugger attach")
ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True)
ptvsd.wait_for_attach()
processors = {
"sst-2": Sst2Processor,
}
output_modes = {
"cola": "classification",
"sst-2": "classification",
#"sts-b": "regression",
}
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')
logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt = '%m/%d/%Y %H:%M:%S',
level = logging.INFO if args.local_rank in [-1, 0] else logging.WARN)
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) and args.do_train:
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)
task_name = args.task_name.lower()
if task_name not in processors:
raise ValueError("Task not found: %s" % (task_name))
processor = processors[task_name]()
output_mode = output_modes[task_name]
label_list = processor.get_labels()
#label_weight = WeightClassCSV(args.data_dir + "/train.csv").get_weights(label_list)
num_labels = len(label_list)
tokenizer = BertTokenizer.from_pretrained(args.bert_model, do_lower_case=args.do_lower_case)
train_examples = None
num_train_optimization_steps = None
if args.do_train:
train_examples = processor.get_train_examples(args.data_dir)
num_train_optimization_steps = int(
len(train_examples) / args.train_batch_size / args.gradient_accumulation_steps) * args.num_train_epochs
if args.local_rank != -1:
num_train_optimization_steps = num_train_optimization_steps // torch.distributed.get_world_size()
# Prepare model
cache_dir = args.cache_dir if args.cache_dir else os.path.join(str(PYTORCH_PRETRAINED_BERT_CACHE), 'distributed_{}'.format(args.local_rank))
model = BertForSequenceClassification.from_pretrained(args.bert_model,
cache_dir=cache_dir,
num_labels=num_labels)
if args.fp16:
model.half()
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
if args.do_train:
param_optimizer = list(model.named_parameters())
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.fp16:
try:
from apex.optimizers import FP16_Optimizer
from apex.optimizers import FusedAdam
except ImportError:
raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use distributed and fp16 training.")
optimizer = FusedAdam(optimizer_grouped_parameters,
lr=args.learning_rate,
bias_correction=False,
max_grad_norm=1.0)
if args.loss_scale == 0:
optimizer = FP16_Optimizer(optimizer, dynamic_loss_scale=True)
else:
optimizer = FP16_Optimizer(optimizer, static_loss_scale=args.loss_scale)
warmup_linear = WarmupLinearSchedule(warmup=args.warmup_proportion,
t_total=num_train_optimization_steps)
else:
optimizer = BertAdam(optimizer_grouped_parameters,
lr=args.learning_rate,
warmup=args.warmup_proportion,
t_total=num_train_optimization_steps)
global_step = 0
nb_tr_steps = 0
tr_loss = 0
if args.do_train:
train_features = convert_examples_to_features(
train_examples, label_list, args.max_seq_length, tokenizer, output_mode)
logger.info("***** Running training *****")
logger.info(" Num examples = %d", len(train_examples))
logger.info(" Batch size = %d", args.train_batch_size)
logger.info(" Num steps = %d", num_train_optimization_steps)
all_input_ids = torch.tensor([f.input_ids for f in train_features], dtype=torch.long)
all_input_mask = torch.tensor([f.input_mask for f in train_features], dtype=torch.long)
all_segment_ids = torch.tensor([f.segment_ids for f in train_features], dtype=torch.long)
if output_mode == "classification":
all_label_ids = torch.tensor([f.label_id for f in train_features], dtype=torch.long)
elif output_mode == "regression":
all_label_ids = torch.tensor([f.label_id for f in train_features], dtype=torch.float)
train_data = TensorDataset(all_input_ids, all_input_mask, all_segment_ids, all_label_ids)
if args.local_rank == -1:
train_sampler = RandomSampler(train_data)
else:
train_sampler = DistributedSampler(train_data)
train_dataloader = DataLoader(train_data, sampler=train_sampler, batch_size=args.train_batch_size, drop_last=True)
model.train()
for _ in trange(int(args.num_train_epochs), desc="Epoch"):
tr_loss = 0
nb_tr_examples, nb_tr_steps = 0, 0
for step, batch in enumerate(tqdm(train_dataloader, desc="Iteration")):
batch = tuple(t.to(device) for t in batch)
input_ids, input_mask, segment_ids, label_ids = batch
# define a new function to compute loss values for both output_modes
logits = model(input_ids, segment_ids, input_mask, labels=None)
if output_mode == "classification":
loss_fct = CrossEntropyLoss()
#loss_fct = CrossEntropyLoss(weight = torch.tensor(label_weight).to(device))
loss = loss_fct(logits.view(-1, num_labels), label_ids.view(-1))
elif output_mode == "regression":
loss_fct = MSELoss()
loss = loss_fct(logits.view(-1), label_ids.view(-1))
if n_gpu > 1:
loss = loss.mean() # mean() to average on multi-gpu.
if args.gradient_accumulation_steps > 1:
loss = loss / args.gradient_accumulation_steps
if args.fp16:
optimizer.backward(loss)
else:
loss.backward()
tr_loss += loss.item()
nb_tr_examples += input_ids.size(0)
nb_tr_steps += 1
if (step + 1) % args.gradient_accumulation_steps == 0:
if args.fp16:
# modify learning rate with special warm up BERT uses
# if args.fp16 is False, BertAdam is used that handles this automatically
lr_this_step = args.learning_rate * warmup_linear.get_lr(global_step, args.warmup_proportion)
for param_group in optimizer.param_groups:
param_group['lr'] = lr_this_step
optimizer.step()
optimizer.zero_grad()
global_step += 1
if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0):
# Save a trained model, configuration and tokenizer
model_to_save = model.module if hasattr(model, 'module') else model # Only save the model it-self
# If we save using the predefined names, we can load using `from_pretrained`
output_model_file = os.path.join(args.output_dir, WEIGHTS_NAME)
output_config_file = os.path.join(args.output_dir, CONFIG_NAME)
torch.save(model_to_save.state_dict(), output_model_file)
model_to_save.config.to_json_file(output_config_file)
tokenizer.save_vocabulary(args.output_dir)
# Load a trained model and vocabulary that you have fine-tuned
model = BertForSequenceClassification.from_pretrained(args.output_dir, num_labels=num_labels)
tokenizer = BertTokenizer.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
else:
model = BertForSequenceClassification.from_pretrained(args.bert_model, num_labels=num_labels)
model.to(device)
train_finish_time = time.time()
train_overall_time = train_finish_time - start_time
if args.do_eval and (args.local_rank == -1 or torch.distributed.get_rank() == 0):
eval_examples = processor.get_dev_examples(args.data_dir)
eval_features = convert_examples_to_features(
eval_examples, label_list, args.max_seq_length, tokenizer, output_mode)
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([f.input_ids for f in eval_features], dtype=torch.long)
all_input_mask = torch.tensor([f.input_mask for f in eval_features], dtype=torch.long)
all_segment_ids = torch.tensor([f.segment_ids for f in eval_features], dtype=torch.long)
if output_mode == "classification":
all_label_ids = torch.tensor([f.label_id for f in eval_features], dtype=torch.long)
elif output_mode == "regression":
all_label_ids = torch.tensor([f.label_id for f in eval_features], dtype=torch.float)
eval_data = TensorDataset(all_input_ids, all_input_mask, all_segment_ids, all_label_ids)
# Run prediction for full data
eval_sampler = SequentialSampler(eval_data)
eval_dataloader = DataLoader(eval_data, sampler=eval_sampler, batch_size=args.eval_batch_size)
model.eval()
eval_loss = 0
nb_eval_steps = 0
preds = []
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():
logits = model(input_ids, segment_ids, input_mask, labels=None)
# create eval loss and other metric required by the task
if output_mode == "classification":
loss_fct = CrossEntropyLoss()
#loss_fct = CrossEntropyLoss(weight = torch.tensor(label_weight).to(device))
tmp_eval_loss = loss_fct(logits.view(-1, num_labels), label_ids.view(-1))
elif output_mode == "regression":
loss_fct = MSELoss()
tmp_eval_loss = loss_fct(logits.view(-1), label_ids.view(-1))
eval_loss += tmp_eval_loss.mean().item()
nb_eval_steps += 1
if len(preds) == 0:
preds.append(logits.detach().cpu().numpy())
else:
preds[0] = np.append(
preds[0], logits.detach().cpu().numpy(), axis=0)
eval_loss = eval_loss / nb_eval_steps
preds = preds[0]
original_preds = preds.copy()
if output_mode == "classification":
preds = np.argmax(preds, axis=1)
elif output_mode == "regression":
preds = np.squeeze(preds)
result = compute_metrics(task_name, preds, all_label_ids.numpy())
auc = compute_auc(original_preds, all_label_ids.numpy())
preds_output_path = os.path.join(args.output_dir, "pred.csv")
write_preds(original_preds, all_label_ids.numpy(), os.path.join(args.data_dir, "dev.csv"), preds_output_path)
result['auc'] = auc
loss = tr_loss/global_step if args.do_train else None
result['eval_loss'] = eval_loss
result['global_step'] = global_step
result['loss'] = loss
result['train_second'] = train_overall_time
result['test_second'] = time.time() - train_finish_time
result['runtime_second'] = time.time() - start_time
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])))
# write to log file
#log_file_path = "./result/bert.log"
log_file_path = args.log_file_path
log_row = []
log_row.append(args.data_dir)
log_row.append(result['precision'])
log_row.append(result['recall'])
log_row.append(result['f1'])
log_row.append(result['train_second'])
log_row.append(result['test_second'])
## append metric report for negative label
#(pre, rec, fs, sup) = precision_recall_fscore_support(y_true=all_label_ids.numpy(), y_pred=preds, pos_label=0, average='binary')
#log_row.append("")
#log_row.append(pre)
#log_row.append(rec)
#log_row.append(fs)
csv_handler.append_row(log_file_path, log_row)
train_duration = train_finish_time - start_time
print("train time is " + str(train_finish_time - start_time))
print("predicting...")
predicted = text_clf.predict(X_dev)
predicted_proba = text_clf.predict_proba(X_dev)
assert (len(predicted_proba) == len(X_dev))
assert (len(X_dev) == len(y_dev))
print("logging...")
(precision, recall, fscore,
support) = metrics.precision_recall_fscore_support(y_dev, predicted)
row = []
row.append(sys.argv[1])
row.append(precision[1])
row.append(recall[1])
row.append(fscore[1])
pos_predicted = transform.map_func(predicted_proba, lambda p: p[1])
auc = metrics.roc_auc_score(y_dev, pos_predicted)
row.append(auc)
accuracy = metrics.accuracy_score(y_dev, predicted)
row.append(accuracy)
csv_handler.append_row(log_file_path, row)
csv_dataset = csv_handler.csv_readlines(csv_input_path)
y_true = transform.map_func(csv_dataset, lambda row: int(row[y_true_col]))
y_pred_score = transform.map_func(csv_dataset,
lambda row: float(row[y_pred_col]))
#y_pred_score = transform.map_func(y_pred_score, lambda score : 1 / (1 + math.exp(-score)))
thred_col = []
if thred_method == "min_max_even":
thred_col = get_threds_by_min_max_even(y_pred_score, num_threds)
else: # sorted score even slot
thred_col = get_threds_by_sorted_score_equal_length(
y_pred_score, num_threds)
if print_header == '1':
csv_handler.append_row(
csv_output_path,
['threshold', 'precision', 'recall', 'fscore', 'support'])
for thred in thred_col:
y_pred = transform.map_func(y_pred_score, lambda score: 1
if score >= thred else 0)
result = precision_recall_fscore_support(y_true, y_pred)
csv_handler.append_row(
csv_output_path,
[thred, result[0][1], result[1][1], result[2][1], result[3][1]])
print("Evaluating ...")
model.eval()
predicts = []
golds = []
with torch.no_grad():
for idx, batch in enumerate(tqdm(test_iter, desc="Iteration")):
inputs, labels = batch.sent, batch.label
inputs = inputs.to(device)
logits = model(inputs)
predict = torch.argmax(logits, dim=1).data.cpu().numpy()
predicts += list(predict)
golds += list(labels.data.cpu().numpy())
precision, recall, f1 = F1(predicts, golds)
print("Precision: %f, Recall: %f, F1: %f" % (precision, recall, f1))
train_time = train_overall_time
test_time = time.time() - train_finish_time
(precision, recall, fscore,
support) = metrics.precision_recall_fscore_support(golds, predicts)
log_row = []
log_row.append(args.dataset)
log_row.append(precision[1])
log_row.append(recall[1])
log_row.append(fscore[1])
log_row.append(train_time)
log_row.append(test_time)
csv_handler.append_row(args.log_file, log_row)