return system absolute path according to relative path, if path dirs not exists and create is True, required folders will be created Args:

evaluating Fact-checking model on data_url Args: model: Fact-checking model data_url: url to test dataset for evaluating, e.g.,

Task 1: Veracity Prediction -- T/F/H... 1) claim + src --> veracity 2) claim + src[oracle_ids] --> veracity 3) cl

convex function:fast->slow, characterizing the easier classification for early phase, and harder classification for later phase. BIG HEAD

return the current date suffix，like '180723.csv' Args: file_type: file type suffix, like '.csv Returns: str: date suffix

return max len

calculating f1 Args: tp: true positive fp: false positive fn: false negative print_result: whether to print resu

obtain domain ids

generate label ids for claims

select oracle ids from ruling_tokenized

dynamic weight average for multi-task learning loss temperature: larger value results in a more even distribution between different tasks.

mode = train, test, val

parsing dict objs to list

for test

1 csv pd.read_csv 2 json return all filenames in dir

Create a summary writer logging to log_dir.

dump python object into json file Args: obj: python object, more information here https://docs.python.org/2/library/json.ht

:param claim_repr: claim tensor (batch_size, n_words, n_hidden) word_repr : src tensor, full text. (2000, n_words, n_hidden)

:param src_tensors: claim_sent, claim_sent_lengths, claim_sent_words, claim_sent_word_lengths, claim_sent_word_indices, claim_sent_nums

generate input tensors for 1 batch： _CLAIM, _CLAIM_TOK, _TGT, _TGT_TOK, _SRC, _SRC_TOK Args: _CLAIM, _CLAIM_TOK: claim sents and its token

choose device Returns: the device specified by name, if name is None, proper device will be returned

generate real index of the tensor. :param len_list: lengths of the tensor. :return: list of real index.

Log a histogram of the tensor of values.

Log a list of images.

@_CLAIM_TOK：list of list

transform list into value-index dict Args: lis: list Returns: value-index dict for lis

load python object form json file Args: json_url: url to json file Returns: python object

collect data with your style

return all filenames in dir

Log a scalar variable.

set random seed for numpy and torch, more information here: https://pytorch.org/docs/stable/notes/randomness.html Args: seed: the

sort a dict by value Args: dic: the dict to be sorted reverse: reverse order or not Returns: sorted dict

Tiles x on dimension dim count times.