majumderb / recipe-personalization Goto Github PK

Hi, I've read your paper. It's very nice and interesting.
I would like to ask you a question about an evaluation of recipe level coherence using an absolute order network.

In your code at recipe-personalization/recipe_gen/analysis/absolute_order_teacher.py L77, the cosine similarity loss is calculated as:

cos_similarity = F.cosine_similarity(repr_1, repr_2, dim=1)

I replicated it and found that the loss value would be negative after training.
Is it acceptable?

Hi, I have got the recipe_gen code from https://github.com/majumderb/recipe-personalization.git and food-com-recipes-and-user-interactions data. How to train and test the code? Do you have any sample code or README with details? Current README says "the code is tested on a Linux server (with NVIDIA GeForce Titan X Pascal / NVIDIA GeForce GTX 1080 Ti) with PyTorch 1.1.0 and Python 3.6."

Hi,
Thanks for your sharing. I have questions to ask you. How is the personalization evaluation calculated. Is it to pre train the evaluation model according to the training set, and then use the pre-trained model to evaluate the personalized performance in the evaluation part. And I can't find the code about this part. Can you tell me which file is related to the evaluation.
Thanks!

There are only raw kaggle data(csv), not preprocessed data(pkl) to train the model.
Is there any addiational code to preprocess the data?

Hi, there a keras version of the model I can look at for better understanding as I'm finding it a bit hard to replicate,

Hi, in your paper you said that you manually construct a list of 58 cooking techniques from 384 cooking actions. Is there any way that I can access these guidelines?
My email address: [email protected]