Sequential learning

This repo contains resources to test a few online convex optimization / best arm identification in bandits algorithms.

Online convex optimization

• Online gradient descent
• Online gradient descend without gradient

Best arm identification in stochastic bandits

Fixed budget

• Uniform sampling
• Successive rejects

Fixed confidence

• UCB-based with heuristic approximation of the GLRT stopping rule
• Uniform sampling with heuristic approximation of the GLRT stopping rule
• TTUCB (Top two with arm drawn by UCB as the leader)
• EB-TC (Top two with empirically best arm as the leader)