Optionally, other data layers can override _get_features_padded_shapes, _get_labels_padded_shapes, _get_features_padding_values, and _get_labels_padding_values, if it is necessary to make batch padding. All subclasses should override _build_features_dataset, that builds input dataset, and _build_labels_dataset, building the labels dataset. batch_size=100.Īll other data layers should subclass it. **kwargs: provides extra name and value params, e.g.the vocabulary file, maximum length, unknown id, labels, etc. init_params: a dictionary containing initialization parameters of the data layer, e.g.It uses to split the text before send it to the vocabulary lookup table. tokenizer: is a function takes as an input a text and tokenises it.The labels_source must be text file(s), each line represents the class that the corresponding sample in the features_source file belongs to. In this case the data layer is supposed to be used in train mode or eval mode. Otherwise, the input pipline will be prepared as labeled data pipeline. labels_source (Optional): If None, the dtat layer only works in the inference mode.Each line in the file represents one sample. Using Invision Community is not always self-explained.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |