search

0xnurl / keras_character_based_ner

Keras Implementation of Character-Based Bi-Directional LSTM RNN for Named Entity Recognition
33 stars 10 forks source link

label shape #2

Closed emresar closed 6 years ago

emresar commented 6 years ago

Hello and thanks for the nice paper and work! I have one question regarding the label shape. You note that the labels should be of shape (batch_size, num_labels), but shouldn't there be a label for each character in an input sentence of length sentence_maxlen? In that case shouldnt the labels be of shape (batch_size, num_labels, sentence_maxlen)?

0xnurl commented 6 years ago

You are correct, the right shape is (batch_size, sentence_maxlen, num_labels). Will fix in documentation. Thanks!