vjg28 / nlp_rl

NLP_RL (Just a test trial)

NLP_RL

Use

• Git clone the repo. Enter datasets folder and unzip all three files. Those files are the preprocessed data of NarrativeQA dataset for the code.
• That just saved you 12 hours of pre-processing. :)
• Run python3 train_span.py
• Wait for everything to happen.

Safety points

• Still don't know whether thr intra-attention function is correct or not. Check Issues in the original repo.
• Multi-gpu code gives error. No clue. :(
• If you have 12 hours to pass your time, do use prep codes with python 2 for preprocessing.
• Still onto testing

Source

• Link

nlp_rl's Issues

Issues with the results.

I got the results as follows
Bleu1=4.75277632252139
Bleu4=0.0002252375950162494
Meteor=2.725085966130152
Rouge=20.281084140962612

Kindly help me out with this.