Steps to reproduce the results

Training and usege of the variant models

Run main_swda_elmo_mean.py for training (set train = True) the models, else it will return the context and non-context models with direct function predict_classes_elmo_mean_features.

Similarly, main_predictor_online.py is for models of ELMo normal features.

They are already called in the following scripts for their usage, and hence before running following scripts, directory params should contain all the weight files and directory features should contain all the elmo features.

Annotating the emotion corpora with dialogue acts

1. To annotate IEMOCAP data: run mocap_dia_act_annotate.py to get context and non-context annotations and save as numpy arrays, and generate final annotation (csv) file.

2. To annotate MELD data: run meld_dia_act_annotate.py to get context and non-context annotations and save as numpy arrays, and generate final annotation (csv) file.

Annotating custom utterances directly

1. Run the ELMO embedding server: running elmo_emb_server.py will start the embedding server.

2. Run DA annotator server: running dia_act_annotator.py will start the annotator server.

3. To annotate your data: run example_api.py script to annotate your own utterances.