Init torch.Tensor Args: tensor: Tensor to be initialized. init_type: Init type, candidate can be found in InitType. low: T

Log warn message msg: Message to log

Log info message msg: Message to log

Calculate precision, recall, fscore Args: standard_count: Standard count predict_count: Predict count righ

For multi-label classification Generate multi-hot for input labels e.g. input: [[0,1], [2]] output: [[1,1,0], [0,0,1]]

Use dict to convert all vocabs to ids

Print dict info

Convert vocab to id. Vocab not in dict map will be map to _UNK

Add key value pair

Calculate the level performance. Args: id_to_label_map: Label id to label name. other_text: Text to judge the other la

Eval the predict result. Args: predicts: Predict probability or predict text label(is_prob is false)

Save the latest evaluation.

Clear all dict

Generate hierarchy label, used in HMCN

Init all dict

Convert label to id. The reason that label is not in label map may be label is filtered or label in validate/test does not occur in train set

Load dict from file. Args: dict_name: Dict name, if None load all dict. Default None. Returns: dict.

Use pretrained embedding to generate dict

Save vocab to file and generate id_to_vocab_dict_map Args: dict_name: Dict name, if None save all dict. Default None.

Convert token to id. Vocab not in dict map will be map to _UNK

Only support hierarchical text classification with BCELoss references: http://www.cse.ust.hk/~yqsong/papers/2018-WWW-Text-GraphCNN.pdf

Tokenization/string cleaning for all datasets except for SST. Original taken from https://github.com/yoonkim/CNN_sentence/blob/master/process

covert the data which text_data are saved as multi-slots, e.g()

Get classification model from configuration

Get data loader: Train, Validate, Test

get parent-children relationships from given hierar_taxonomy hierar_taxonomy: parent_label \t child_label_0 \t child_label_1 \n

input texts should be json objects

Save confusion matrix Args: file_name: File to save to. confusion_matrix: Confusion Matrix. Returns:

Save precision, recall, fscore Args: file_name: File to save to. precision_category: Precision dict. recal

Performs a single optimization step. Arguments: closure (callable, optional): A closure that reevaluates the model

Update lr

Another way to do this is keep the file handler. But when DataLoader's num_worker bigger than 1, error will occur. Args: