some issues in data processing about mmd

[shubhamagarwal92]

1. Thanks for noting that. I will update the $DATA_DIR path.
2. You can provide the config version when prompted. I will also update the default version. (I can also provide you other config versions that I tried.)

Let me know if you face any other issues or want me to upload the data

[khoaipx]

If I want to re-produce the baseline result (in the paper Amrita Saha et. al., Towards Building Large Scale Multimodal Domain-Aware Conversation Systems, 2017), please tell me how to config.

[shubhamagarwal92]

As we noted in the paper, as well as in this issue, we are not able to achieve BLEU of 56.67 that they report in the paper. With their config, dataset and code, we got a BLEU of 37.67.

[shubhamagarwal92]

Hi @khoaipx

Please see the commit 503b481.

I have uploaded the data that I used here. It contains the versions for both context 2 and context 5 and authors (Shah et al. version). Unzip and put in data folder. Please take care of the file paths.

Also, I have uploaded sample config to run our code on their version of data.

[shubhamagarwal92]

Closing for now. Reopen if you face any other issues. I will also appreciate if you have PRs related to issues with file path. :)