Recommender Systems using Nvidia Merlin

We're going to explore Recommender Systems using Nvidia's opensource framework for recsys Merlin.

We need to install all the dependencies.

As we’re going to experiment, we’ll keep this build setup “big” and unoptimised.

First of all, we’ll be doing CUDA computations so we’re going to use anaconda for virtual development setup.

We install Pytorch with the official guidelines from pytorch.org

conda install pytorch torchvision torchaudio
 pytorch-cuda=11.7 -c pytorch -c nvidia

Once this is done, we should proceed installing usual dev dependencies.

conda install jupyter pandas matplotlib

And finally we install the Merlin library packages, with the instructions from Nvidia Merlin Github

conda install -c nvidia -c rapidsai -c numba /
 -c conda-forge merlin-core python=3.7 cudatoolkit=11.2

Next we have to install NVTabular, which is a Python library also from Nvidia that handles tabular data on the GPU, as part of RAPIDS framework.

conda install -c nvidia -c rapidsai -c numba /
 -c conda-forge nvtabular python=3.7 cudatoolkit=11.2

1 Getting The data