DMN+ implementation in Pytorch for question answering on the bAbI 10k dataset.
Input Module for DMNPlus
Run the included shell script to fetch the data :
$ chmod +x fetch_data.sh
$ ./fetch_data.sh$ python train.py| Task ID | This Repo | Xiong et al | Task ID | This Repo | Xiong et al |
|---|---|---|---|---|---|
| 1: single-supporting-fact | 100% | 100% | 11: basic-coreference | 100% | 100% |
| 2: two-supporting-facts | 95.6% | 99.7% | 12: conjunction | 100% | 100% |
| 3: three-supporting-facts | 85.0% | 98.9% | 13: compound-coreference | 100% | 100% |
| 4: two-arg-relations | 100% | 100% | 14: time-reasoning | 98.6% | 99.8% |
| 5: three-arg-relations | 99.5% | 99.5% | 15: basic-deduction | 100% | 100% |
| 6: yes-no-questions | 100% | 100% | 16: basic-induction | 48.0% | 54.7% |
| 7: counting | 98.8% | 97.6% | 17: positional-reasoning | 87.9% | 95.8% |
| 8: lists-sets | 100% | 100% | 18: size-reasoning | 97.0% | 97.9% |
| 9: simple-negation | 100% | 100% | 19: path-finding | 99.7% | 100% |
| 10: indefinite-knowledge | 100% | 100% | 20: agents-motivations | 100% | 100% |
| Mean | 95.45% | 97.20% |
Download pre-trained DMN+ Model then run:
$ python demo.py