Book Recommender System
a project using ALS matrix factorization and SGD
Alternating least square matrix factorization is a simple algorithm which anybody who has taken a linear algebra course can understand. Although writing the implementation is PyTorch was easy, the biggest obstacles were the huge size of the matrix, sparsity of the matrix, and hyperparameter tuning cost.
EDA and methods
Hyperparameter Tuning
I did batch gradient descent and stored the matrix as tensors for sake of memory. Also, I experimented with lots of gradient descent techniques like SGD, ADAM, AdaGrad, etc.