A Contextual-Bandit-Based Approach for Informed Decision-Making in Clinical Trials

. 2022 Aug 21;12(8):1277. doi: 10.3390/life12081277

Algorithm 1 Thompson sampling

1:
for $i = 1, 2, \dots,$ do
2:
for $u \in {1, \dots, K}$ do
3:
Sample $\hat{θ_{u}} \sim b e t a (α_{u}, β_{u})$ ▹ sample model
$U_{i} = {arg max}_{u} \hat{θ_{u}}$ ▹ select and apply action
4:
Apply $U_{i}$ and observe $C_{i}$
5:
$(α_{U_{i}}, β_{U_{i}}) \leftarrow (α_{U_{i}}, β_{U_{i}}) + (C_{i}, 1 - C_{i})$ ▹ update distribution