Hi I am currently facing an issue when trying to run a large number of Lasso Regressions on the same data. I have a dataset with multiple 100k rows and 365 columns (data of one year). What I need to do is to fit and predict one Lasso Regression per Row based on all other rows so that I end up with as many individual regressions as rows.
I struggle to find a way other than a loop to execute this in a rather performant way. I tried to experiment with Python's joblib package, which increased the performance, but I am still looking for a faster way if there is any. I also tried to find a way to vectorize this problem, but I did not find a solution. I also glanced at keras to implement the Lasso regression as an aNN.
I can hardly imagine that it is uncommon to fit a large number of simple models for specific tasks. So my question is: What is the best approach to reduce execution time of this problem statement? From my understanding, this problem can not be optimized with the use of GPUs as lots of individual models need to be fitted and therefore the CPU will be the bottleneck regardless.
Btw: I can get access to a GPU and I have a CPU with 16 cores.