/dialogue-hred-vhred

HRED VHRED VHCR for Multi-Turn Dialogue Systems

Primary LanguagePython

HRED VHRED VHCR for Multi-Turn Dialogue Systems

Preprocess data

./data: train.txt, dev.txt, test.txt

format:

u1 </s> u2 </s> u3 \t response

example:

w11 w12 w13 </s> w21 w22 </s> w31 w32 w33 w34 \t w1 w2 w3

then:

python prepare_data.py

Training

Go to the model directory and set the save_dir in configs.py (this is where the model checkpoints will be saved)

We provide our implementation of VHCR, as well as our reference implementations for HRED and VHRED.

To run training:

python train.py --model=<model> --batch_size=<batch_size>

For example:

  1. Train HRED:
python train.py  --model=HRED
  1. Train VHRED with word drop of ratio 0.25 and kl annealing iterations 250000:
python train.py --model=VHRED --batch_size=40 --word_drop=0.25 --kl_annealing_iter=250000
  1. Train VHCR with utterance drop of ratio 0.25:
python train.py --model=VHCR --batch_size=40 --sentence_drop=0.25 --kl_annealing_iter=250000

Evaluation

To evaluate the word perplexity:

python eval.py --model=<model> --checkpoint=<path_to_your_checkpoint>

For embedding based metrics, you need to download Google News word vectors, unzip it and put it under the datasets folder. Then run:

python eval_embed.py --model=<model> --checkpoint=<path_to_your_checkpoint>

Generation

To generate the response for the test set:

python test.py --model=<model> --checkpoint=<path_to_your_checkpoint>

BLEU and DIST

python metrics.py

Reference