Building Recommender Systems Faster Using Jupyter Notebooks from NGC

Originally published at: Building Recommender Systems Faster Using Jupyter Notebooks from NGC | NVIDIA Developer Blog

The NVIDIA NGC team is hosting a webinar with live Q&A to dive into this Jupyter notebook available from the NGC catalog. Learn how to use these resources to kickstart your AI journey. Register now: NVIDIA NGC Jupyter Notebook Day: Recommender System. Recommender systems deal with predicting user preferences for products based on historical behavior…

Thanks for the blog post! I followed along until I couldn’t proceed further because I’m using a Pascal GPU at home (1080Ti).
A couple suggestions:

  • List the supported GPU architectures on the blog post (Volta, Turing and Ampere)
  • Check the order and content of the steps to set up the dataset against the quick start guide posted here - NVIDIA NGC

Thanks, looking forward to the webinar!

1 Like

Update: I was able to get past the CUDA error in the training step and it’s training on my lowly 1080Ti! :)

@daqieq - would you share how you resolved CUDA error on your 1080Ti? Thanks.

@lz3awuo - to be honest … I just ran it again to check it before I gave up and shut down.

Thanks for the article. I have a problem with the premise of your article. And that would be this statement: “The Variational Autoencoder (VAE) shown here is an optimized implementation of the architecture first described in Variational Autoencoders for Collaborative Filtering and can be used for recommendation tasks.”

And to be clear, with the very last part of the sentence: “VAE can be used for recommendation tasks”. I’m not sure if that’s true. I mean there might have been attempts to use a VAE for such a purpose but I don’t think that works. A recommender system requires a space where distance represents similarity. And I don’t see how the latent space of a VAE satisfies that requirement.

I appreciate it if you could help me understand how the latent space of a VAE is suitable for a recommender system.

Thanks.

@mehranziadloo. The demo and this blog use this nvidia/ngc resource: NVIDIA NGC and in this resource we trained the model with movie rating dataset so the model can estimate the rate of a movie for the new user. With a trained model, you can run inference to predict what items is a new user most likely to interact with so when we trained with rating dataset it can recommend a movie based on the estimated rate.