This is forked from Qipeng Guo's repository, and excludes large data files.

GraphWriter-DGL

In this example we implement the GraphWriter, Text Generation from Knowledge Graphs with Graph Transformers in DGL. And the author's code.

Dependencies

PyTorch >= 1.2
tqdm
pycoco

Usage

  sh run.sh
  sh test.sh

Result on AGENDA

	BLEU	METEOR	training time per epoch
paper	14.3+-1.01	18.8+-0.28	1970s
this repo	14.31+-0.34	19.74+-0.69	1192s

We use the author's code for the speed test, and our testbed is V100 GPU.

	BLEU	detok BLEU	METEOR
this repo, greedy, two layers	13.97 +- 0.40	13.78 +- 0.46	18.76 +- 0.36
this repo, beam 4, length penalty 1.0, two layers	14.66 +- 0.65	14.53 +- 0.52	19.50 +- 0.49
this repo, beam 4, length penalty 0.0, two layers	14.33 +- 0.39	14.09 +- 0.39	18.63 +- 0.52
this repo, greedy, six layers	14.17 +- 0.46	14.01 +- 0.51	19.18 +- 0.49
this repo, beam 4, length penalty 1.0, six layers	14.31 +- 0.34	14.35 +- 0.36	19.74 +- 0.69
this repo, beam 4, length penalty 0.0, six layers	14.40 +- 0.85	14.15 +- 0.84	18.86 +- 0.78

We repeat the experiment five times.