royeis / FlowNLG

FlowNLG

Enriched variant of DeepNLG, introducing information flow hints.

Prerequistes

It is recommended to run the code in a docker container or a designated conda envronment.

• clone the repository.
• install requirements with pip install -r requirements.txt. (possibly change torch version depending on your version of cuda )

The code is compatible with cuda, and it is recommended to run it on a cuda machine, but runs on cpu as well.

Generating enriched data from DeepNLG

Enriched dataset is located in data folder by default. Running DatasetGenerator.py will create the enriched dataset from DeepNLG located in DeepNLG_data

Evaluating text generated from FlowNLG entries

To generate text from FlowNLG and measure BLEU metric with respect to references run EvalOracle.py