Skip to content

Commit

Permalink
bool arguments fix
Browse files Browse the repository at this point in the history
  • Loading branch information
@saransh-mehta
saransh-mehta committed Jun 13, 2020
1 parent 889f978 commit 2a6db66
Show file tree
Hide file tree
Showing 10 changed files with 45 additions and 82 deletions.
8 changes: 4 additions & 4 deletions examples/answerability_detection/answerability_detection_msmarco.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -141,11 +141,11 @@
" --eval_batch_size 16 \\\n",
" --grad_accumulation_steps 2 \\\n",
" --log_per_updates 250 \\\n",
" --save_per_updates 16000 \\\n",
" --eval_while_train True \\\n",
" --test_while_train True \\\n",
" --max_seq_len 324 \\\n",
" --silent True "
" --save_per_updates 16000 \\\n",
" --eval_while_train \\\n",
" --test_while_train \\\n",
" --silent"
]
},
{
Expand Down
6 changes: 3 additions & 3 deletions examples/entailment_detection/entailment_snli.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -139,10 +139,10 @@
" --eval_batch_size 64 \\\n",
" --grad_accumulation_steps 1 \\\n",
" --log_per_updates 100 \\\n",
" --eval_while_train True \\\n",
" --test_while_train True \\\n",
" --max_seq_len 128 \\\n",
" --silent True "
" --eval_while_train \\\n",
" --test_while_train \\\n",
" --silent"
]
},
{
Expand Down
7 changes: 3 additions & 4 deletions examples/intent_ner_fragment/intent_ner_fragment.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -168,7 +168,6 @@
"metadata": {},
"outputs": [],
"source": [
"\n",
"!python ../../train.py \\\n",
" --data_dir '../../data/bert-base-uncased_prepared_data' \\\n",
" --task_file 'tasks_file_snips.yml' \\\n",
Expand All @@ -178,10 +177,10 @@
" --eval_batch_size 32 \\\n",
" --grad_accumulation_steps 2 \\\n",
" --log_per_updates 50 \\\n",
" --eval_while_train True \\\n",
" --test_while_train True \\\n",
" --max_seq_len 50 \\\n",
" --silent True "
" --eval_while_train \\\n",
" --test_while_train \\\n",
" --silent "
]
},
{
Expand Down
6 changes: 3 additions & 3 deletions examples/ner_pos_tagging/ner_pos_tagging_conll.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,10 +58,10 @@
" --eval_batch_size 32 \\\n",
" --grad_accumulation_steps 1 \\\n",
" --log_per_updates 50 \\\n",
" --eval_while_train True \\\n",
" --test_while_train True \\\n",
" --max_seq_len 50 \\\n",
" --silent True "
" --eval_while_train \\\n",
" --test_while_train \\\n",
" --silent"
]
},
{
Expand Down
6 changes: 3 additions & 3 deletions examples/query_correctness/query_correctness.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,10 +58,10 @@
" --eval_batch_size 32 \\\n",
" --grad_accumulation_steps 1 \\\n",
" --log_per_updates 20 \\\n",
" --eval_while_train True \\\n",
" --test_while_train True \\\n",
" --max_seq_len 50 \\\n",
" --silent True"
" --eval_while_train \\\n",
" --test_while_train \\\n",
" --silent"
]
},
{
Expand Down
4 changes: 2 additions & 2 deletions examples/query_pair_similarity/query_similarity_qqp.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -96,9 +96,9 @@
" --eval_batch_size 16 \\\n",
" --grad_accumulation_steps 2 \\\n",
" --log_per_updates 50 \\\n",
" --eval_while_train True \\\n",
" --test_while_train True \\\n",
" --max_seq_len 200 \\\n",
" --eval_while_train \\\n",
" --test_while_train \\\n",
" --silent True "
]
},
Expand Down
6 changes: 3 additions & 3 deletions examples/query_type_detection/query_type_detection.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -59,10 +59,10 @@
" --eval_batch_size 64 \\\n",
" --grad_accumulation_steps 1 \\\n",
" --log_per_updates 100 \\\n",
" --eval_while_train True \\\n",
" --test_while_train True \\\n",
" --max_seq_len 60 \\\n",
" --silent True"
" --eval_while_train \\\n",
" --test_while_train \\\n",
" --silent"
]
},
{
Expand Down
5 changes: 2 additions & 3 deletions models/model.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -225,9 +225,8 @@ def update_step(self, batchMetaData, batchData):
if self.lossClassList[taskId] and (target is not None):
self.taskLoss = self.lossClassList[taskId](logits, target, attnMasks=modelInputs[2])
#tensorboard details
if self.params['tensorboard']:
self.tbTaskId = taskId
self.tbTaskLoss = self.taskLoss.item()
self.tbTaskId = taskId
self.tbTaskLoss = self.taskLoss.item()
taskLoss = self.taskLoss / self.params['grad_accumulation_steps']
taskLoss.backward()
self.accumulatedStep += 1
Expand Down
48 changes: 22 additions & 26 deletions train.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -29,51 +29,48 @@ def make_arguments(parser):
help = 'path to save the model')
parser.add_argument('--epochs', type = int, required=True,
help = 'number of epochs to train')
parser.add_argument('--finetune', type = bool, default= False,
help = "If only the shared model is to be loaded with saved pre-trained multi-task model.\
In this case, you can specify your own tasks with task file and use the pre-trained shared model\
to finetune upon.")
parser.add_argument('--freeze_shared_model', type = bool, default=False,
parser.add_argument('--freeze_shared_model', default=False, action='store_true',
help = "True to freeze the loaded pre-trained shared model and only finetune task specific headers")
parser.add_argument('--train_batch_size', type = int, default=8,
help='batch size to use for training')
parser.add_argument('--eval_batch_size', type = int, default = 32,
help = "batch size to use during evaluation")
parser.add_argument('--eval_while_train', type = bool, default= True,
help = "if evaluation on dev set is required during training.")
parser.add_argument('--test_while_train', type = bool, default = True,
help = "if evaluation on test set is required during training.")
parser.add_argument('--grad_accumulation_steps', type =int, default = 1,
help = "number of steps to accumulate gradients before update")
parser.add_argument('--num_of_warmup_steps', type=int, default = 0,
help = "warm-up value for scheduler")
parser.add_argument('--grad_clip_value', type = float, default=1.0,
help = "gradient clipping value to avoid gradient overflowing" )
parser.add_argument('--debug_mode', default = False, type = bool,
help = "record logs for debugging if True")
parser.add_argument('--log_file', default='multi_task_logs.log', type = str,
help = "name of log file to store")
parser.add_argument('--log_per_updates', default = 10, type = int,
help = "number of steps after which to log loss")
parser.add_argument('--silent', type = bool, default = True,
help = "Only write logs to file if True")
parser.add_argument('--seed', default=42, type = int,
help = "seed to set for modules")
parser.add_argument('--max_seq_len', default=128, type =int,
help = "max seq length used for model at time of data preparation")
parser.add_argument('--tensorboard', default=True, type = bool,
help = "To create tensorboard logs")
parser.add_argument('--save_per_updates', default = 0, type = int,
help = "to keep saving model after this number of updates")
parser.add_argument('--limit_save', default = 10, type = int,
help = "max number recent checkpoints to keep saved")
parser.add_argument('--load_saved_model', type=str, default=None,
help="path to the saved model in case of loading from saved")
parser.add_argument('--resume_train', type=bool, default=False,
help="True for resuming training from a saved model")
parser.add_argument('--eval_while_train', default = False, action = 'store_true',
help = "if evaluation on dev set is required during training.")
parser.add_argument('--test_while_train', default=False, action = 'store_true',
help = "if evaluation on test set is required during training.")
parser.add_argument('--resume_train', default=False, action = 'store_true',
help="Set for resuming training from a saved model")
parser.add_argument('--finetune', default= False, action = 'store_true',
help = "If only the shared model is to be loaded with saved pre-trained multi-task model.\
In this case, you can specify your own tasks with task file and use the pre-trained shared model\
to finetune upon.")
parser.add_argument('--debug_mode', default = False, action = 'store_true',
help = "record logs for debugging if True")
parser.add_argument('--silent', default = False, action = 'store_true',
help = "Only write logs to file if True")
return parser


parser = argparse.ArgumentParser()
parser = make_arguments(parser)
args = parser.parse_args()
Expand Down Expand Up @@ -193,9 +190,8 @@ def main():
allParams['gpu'] = torch.cuda.is_available()
logger.info('task parameters:\n {}'.format(taskParams.taskDetails))

if args.tensorboard:
tensorboard = SummaryWriter(log_dir = os.path.join(logDir, 'tb_logs'))
logger.info("Tensorboard writing at {}".format(os.path.join(logDir, 'tb_logs')))
tensorboard = SummaryWriter(log_dir = os.path.join(logDir, 'tb_logs'))
logger.info("Tensorboard writing at {}".format(os.path.join(logDir, 'tb_logs')))

# making handlers for train
logger.info("Creating data handlers for training...")
Expand Down Expand Up @@ -268,11 +264,11 @@ def main():
taskName,
avgLoss,
model.taskLoss.item()))
if args.tensorboard:
tensorboard.add_scalar('train/avg_loss', avgLoss, global_step= model.globalStep)
tensorboard.add_scalar('train/{}_loss'.format(taskName),
model.taskLoss.item(),
global_step=model.globalStep)

tensorboard.add_scalar('train/avg_loss', avgLoss, global_step= model.globalStep)
tensorboard.add_scalar('train/{}_loss'.format(taskName),
model.taskLoss.item(),
global_step=model.globalStep)

if args.save_per_updates > 0 and ( (model.globalStep+1) % args.save_per_updates)==0 and (model.accumulatedStep+1==args.grad_accumulation_steps):
savePath = os.path.join(args.out_dir, 'multi_task_model_{}_{}.pt'.format(epoch,
Expand Down
31 changes: 0 additions & 31 deletions utils/tranform_functions.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -672,37 +672,6 @@ def msmarco_answerability_detection_to_tsv(dataDir, readFile, wrtDir, transParam
devDf.to_csv(os.path.join(wrtDir, 'msmarco_answerability_test.tsv'), sep='\t', index=False, header=False)
print('Test file written at: ', os.path.join(wrtDir, 'msmarco_answerability_test.tsv'))

def query_correctness_to_tsv(dataDir, readFile, wrtDir, transParamDict, isTrainFile=False):

"""
- Query correctness transformed file
For using this transform function, set ``transform_func`` : **query_correctness_to_tsv** in transform file.
Args:
dataDir (:obj:`str`) : Path to the directory where the raw data files to be read are present..
readFile (:obj:`str`) : This is the file which is currently being read and transformed by the function.
wrtDir (:obj:`str`) : Path to the directory where to save the transformed tsv files.
transParamDict (:obj:`dict`, defaults to :obj:`None`): Dictionary of function specific parameters. Not required for this transformation function.
"""
print('Making data from file {}'.format(readFile))
df = pd.read_csv(os.path.join(dataDir, readFile), sep='\t', header=None, names = ['query', 'label'])

# we consider anything above 0.6 as structured query (3 or more annotations as structured), and others as non-structured

#df['label'] = [str(lab) for lab in df['label']]
df['label'] = [int(lab>=0.6)for lab in df['label']]

data = [ [str(i), str(row['label']), row['query'] ] for i, row in df.iterrows()]

wrtDf = pd.DataFrame(data, columns = ['uid', 'label', 'query'])

#writing
wrtDf.to_csv(os.path.join(wrtDir, 'query_correctness_{}'.format(readFile)), sep="\t", index=False, header=False)
print('File saved at: ', os.path.join(wrtDir, 'query_correctness_{}'.format(readFile)))

def clinc_out_of_scope_to_tsv(dataDir, readFile, wrtDir, transParamDict, isTrainFile=False):

"""
Expand Down

0 comments on commit 2a6db66

Please sign in to comment.