Regarding the experimental results and the generation of files such as ans_tokenizer

[3514000142]

Hello, I'm trying to reproduce your experiment. I have some questions that I don't understand,
(1) Such as how the three files ans_list.pkl,ans_tokenizer.pt and an_tokenizer.pkl in the external folder are generated and how I should operate. I saw someone ask this question. But I still don't understand.
(2) Also, is the Abnormalitry.csv file in the medvqa-2019 dataset not used? If I use this file, I need to update the three files in external_data in the above question. The following is the result I reproduced without adding the Abnormalitry.csv file. It's a bit low. I downloaded the dataset in MA3E, but the result of the close class problem in my results is 0. Why is that?Should I choose (vga/test/score) as the result or (vga/test/score best epoch) as the result.

Looking forward to your reply

[3514000142]

[GaryGuTC]

Hi

Such as how the three files ans_list.pkl,ans_tokenizer.pt and an_tokenizer.pkl in the external folder are generated and how I should operate. I saw someone ask this question. But I still don't understand.

Due to the long time, I can not remember the whole process. Please refer here first. Which steps are you confused about?

Also, is the Abnormalitry.csv file in the medvqa-2019 dataset not used? If I use this file, I need to update the three files in external_data in the above question. The following is the result I reproduced without adding the Abnormalitry.csv file. It's a bit low. I downloaded the dataset in MA3E, but the result of the close class problem in my results is 0. Why is that?Should I choose (vga/test/score) as the result or (vga/test/score best epoch) as the result.

1. Datasets: Please refer here to download and check dataset.
2. Results: Please ensure the environment is the same as ours.
3. The best epoch will be fine.