The Summarizer environment is a domain for text generation in the context of summarization.
Generate summaries by extracting sentences from reference documents.
- Download CNN-DailyMail finished file binaries;
- Use gym_summarizer.utils.BatchCNNDMLoader to precompute embeddings (WARNING: precomputing will take a very long time and the resulting files will occupy >30Gb of storage);
- Use run_experiment.py and evaluate_model.py to train agents and evaluate them.