Regarding the issue of non-existent dataset/Amazon toys regen/toy/inter. csv

[ComeFromTheMars]

Regarding the seventh step of running the DR4SR training in README.md, I ran the reference command you provided, but it prompted me that I don't have this folder. Did this file be generated in which step? Or do I need to create a folder myself and put the files generated by the inference into it?

[ShadowTinker]

Hi, @ComeFromTheMars , thanks for your attention to our work!

We have fixed this bug. You can set the train_file entry to '_regen' in configs/amazon-toys.yaml to train a model with the regenerated dataset. And you can set the entry to '_ori' to train a model with the original dataset.

[ComeFromTheMars]

Hi, @ComeFromTheMars , thanks for your attention to our work!

We have fixed this bug. You can set the train_file entry to '_regen' in configs/amazon-toys.yaml to train a model with the regenerated dataset. And you can set the entry to '_ori' to train a model with the original dataset.

Thank you very much for your reply. Additionally, I have another question regarding the generation model you mentioned in the article, which has a parameter for balancing and restricting generation modes. Can this parameter be set?

[ShadowTinker]

Hi, @ComeFromTheMars , thanks for your attention to our work!
We have fixed this bug. You can set the train_file entry to '_regen' in configs/amazon-toys.yaml to train a model with the regenerated dataset. And you can set the entry to '_ori' to train a model with the original dataset.

Thank you very much for your reply. Additionally, I have another question regarding the generation model you mentioned in the article, which has a parameter for balancing and restricting generation modes. Can this parameter be set?

Sorry for the late reply. This parameter can be set in the following code:

DR4SR/3.Hybrid_inference.py

Lines 197 to 200 in 67aa7ad

	if random.random() > 1 or i <= 1:
	prob = inference_mask(prob, src, ys)
	else:
	prob = inference_mask_generative(prob, src, ys)

For example, you can change it to if random.random() > 0.3 or i <= 1: , and it will choose the generative mode with probability 30%.

[ComeFromTheMars]

Hi, @ComeFromTheMars , thanks for your attention to our work!
We have fixed this bug. You can set the train_file entry to '_regen' in configs/amazon-toys.yaml to train a model with the regenerated dataset. And you can set the entry to '_ori' to train a model with the original dataset.

Thank you very much for your reply. Additionally, I have another question regarding the generation model you mentioned in the article, which has a parameter for balancing and restricting generation modes. Can this parameter be set?

Sorry for the late reply. This parameter can be set in the following code:

DR4SR/3.Hybrid_inference.py

Lines 197 to 200 in 67aa7ad

	if random.random() > 1 or i <= 1:
	prob = inference_mask(prob, src, ys)
	else:
	prob = inference_mask_generative(prob, src, ys)

For example, you can change it to if random.random() > 0.3 or i <= 1: , and it will choose the generative mode with probability 30%.

Thank you very much for your reply！