khanhnamle1994 / metarec

PyTorch Implementations For A Series Of Deep Learning-Based Recommendation Models

Hi, I'm James Le 👋🏼 👨🏽‍🎓 👨🏽‍💻 🇻🇳 🇺🇸

I'm a machine learning researcher who is deeply interested in tinkering with neural networks, building recommendation systems, and gathering best practices of machine learning in production. I write often about theoretical deep learning and applied MLOps.

• 🔭 I’m doing research at the intersection of deep learning and recommendation systems.
• 🌱 I’m cracking the data science and machine learning engineering interviews.
• 👯 I’m looking to evangelize the Full Stack Deep Learning community.
• 🤔 I’m looking for help to get more guests on Datacast.

Find me around the web 🌎:

• Everything about me in my website 🖥
• Writing in public on Medium ✍🏽
• Interviewing people on Datacast 🎤
• Sharing updates on LinkedIn and Twitter 💼 🐦

metarec's Issues

How to compute the precision and recall for MF with side information.

Dear Khan,
how to compute the precision and recall along with MSE
for matrix factorization with side information.

waiting for your answer.

the validation of NVIDIA-DeepRec model

in DeepRec implementation, when you've trained the model why you've specified that: validation_data=[users_items_matrix_train_zero, users_items_matrix_validate] and not: [users_items_matrix_train_validate, users_items_matrix_validate]?

Doubt regarding use of other datasets

Hi there,
I wanted to run the CDAE, ESAE and SVAE models for other datasets ( Movielens 20M/10M /Netflix /custom datset having UserId, ItemId, rating).

Could you guide me regarding the data preprocessing and the directory set up?
Is there a .py file that does the mentioned ? If, yes what changes do I have to make and how do I go forward with it ?

Thanks in advance!

Inappropriate parameters for pandas preprocessing data

https://github.com/khanhnamle1994/MetaRec/blob/master/Autoencoders-Experiments/CDAE-PyTorch/DataUtils.py
at line 55, 56

num_items_by_user = data.groupby('user', as_index=False).size()
num_users_by_item = data.groupby('item', as_index=False).size()

Does the as_index should be set to True because the later convert to pandas dataframe user_frame = num_items_by_user.to_frame() and assign columns name user_frame.columns = ['item_cnt'] as only as_index=True will result in a pandas Series, otherwise it already be a dataframe so cannot be converted?

update of weights in DeepRec

when training an autoencoder for rating based collaborative filtering, the weights that are associated to unrated items are not updated, does your custom loss function in DeepRec implementation (masked se) provide that?