HMM的训练，即根据训练语料对模型参数进行估计, 因为我们有观测序列以及其对应的状态序列，所以我们 可以使用极大似然估计的方法来估计隐马尔可夫模型的参数 参数: word_lists: 列表，其中每个元素由字组成的列

:param seq: :param id2label: :param markup: :return:

Creates examples for the training and dev sets.

Creates examples for the training and dev sets.

Creates examples for the training and dev sets.

Creates examples for the training and dev sets.

读取数据

计算混淆矩阵

将结果用表格的形式打印出来，像这个样子： precision recall f1-score support B-LOC 0.775 0.757 0.766 1084

Convert sparse matrix to tuple representation.

Serializes this instance to a JSON string.

Resets wait counter and cooldown counter.

Create dataloader. Args: dataset(obj:`paddle.io.Dataset`): Dataset instance. mode(obj:`str`, optional, defaults to obj:`train

map ori offset to token offset

Symmetrically normalize adjacency matrix.

Saves model when validation loss decreases.

用于保存模型

设置整个开发环境的seed :param seed: :param device: :return:

抽取序列特征

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

labels_paths: [[],[],[],....] pred_paths: [[],[],[],.....] :param label_paths: :param pred_paths: :return:

output1 - True - algorithm for matrix, False - vector; output2 - need reshape

Used to mask padded positions: creates a (batch_size, max_len) boolean mask from a tensor of sequence lengths, where 1 means keep ele

test single model first, load model from folder and do prediction

Threshold value could be adjusted to shrink eta_min and eta_max values.