You want to solve Lasso regression but know only Gradient Descent and can't compute gradient of lasso regularizer.
- Proximal Gradient (PG) descent
- Accelerated Proximal Gradient (APG) descent (Nesterov's)
- Accelerated Proximal Gradient with Restart (APGRestart) descent
- Coordinate descent (CD)
- Running delta: instead of computing fresh delta at every iteration, compute incrementally by difference.
- matrix.dot(sparse_vector): filter out zero rows of the vector