Return the metrics of precision, recall and f-score, based on the number (We make this small piece of function in order to reduce the code ef

Decode the batch input :param batchInput: :return:

Calculate the log_sum_exp trick for the tensor. :param vec: [batchSize * from_label * to_label]. :return: [batchSize * to_label]

Backward algorithm. A benchmark implementation which is ready to use. :param lstm_scores: shape: (batch_size, sent_len, label_size) N

Evaluate a batch of instances and handling the padding positions. :param batch_insts: a batched of instances. :param batch_pred_ids: Sha

Copying the optimizer code from HuggingFace.

Build the mapping from label to index and index to labels. :param insts: list of instances. :return:

Use IOBES tagging schema to replace the IOB tagging schema in the instance :param insts: :return:

Load the elmo vectors and the vector will be saved within each instance with a member `elmo_vec` :param file: the vector files for the ELMo v

Returns the space that contains all the instances that has this label.

Returns the search space of instances whose labels are supersets of instance inst.

Get the bitmask state of an instance based on the labels it has.

Calculate all scores by adding up the transition scores and emissions (from lstm). Basically, compute the scores for each edges betwe

Build a search space for each possible combination of labels using bitmask. Note that for a combination x, we choose the ins

build the embedding table with pretrained word embeddings (if given otherwise, use random embeddings) :return:

Build a search space for each label, which contains all the instances that have this label. An instance may appear in multipl

Build the vocab 2 idx for all instances :param train_insts: :param dev_insts: :param test_insts: :return:

Note: This function is not used unless you want to compute the marginal probability Forward-backward algorithm to compute the margina

Calculate the scores for the gold instances. :param all_scores: (batch, seq_len, label_size, label_size) :param word_seq_lens

Calculate the scores with the forward algorithm. Basically calculating the normalization term :param all_scores: (batch_size x max_se

Path to result txt file

Method to decay the learning rate :param config: configuration :param optimizer: optimizer :param epoch: epoch number :return:

Configure optional arguments for the command line argument parser.

Configure optional arguments for the command line argument parser.

Configure optional arguments for the command line argument parser.

Configure optional arguments for the command line argument parser.

sents = [['word_a', 'word_b'], ['word_aaa', 'word_bccc', 'word_ccc']]

sents = [['word_a', 'word_b'], ['word_aaa', 'word_bccc', 'word_ccc']]

Read the pretrained word embeddings, return the complete embeddings and the embedding dimension :return:

Use viterbi to decode the instances given the scores and transition parameters :param all_scores: (batch_size x max_seq_len x num_lab

model_archived_file: ends with "tar.gz" OR directly use the model folder patth

This word embedder allows to static contextualized representation. :param config: :param print_info:

sents: we use sentences if we want to build dataset from sentences directly instead of file

Read the dataset into Instance

sents: we use sentences if we want to build dataset from sentences directly instead of file