robinvanemden
Robin van Emden
Recently Published
LinUCB LinUCB 80% scarsity plot
80% scarse, that is, 20% of the data is used to learn
LinUCB LinUCB 0% scarsity plot
0% scarse, that is, 100% of the data is used to learn
LinUCB 99% scarsity plot
99% scarse, that is, 1% of the data is used to learn
Disjoint LinUCB vs EpsilonGreedy vs Random on Yahoo Data 2
simulations <- 1
horizon <- 10000000
bandit <- YahooBandit$new(k = 217L, d = 6L, arm_lookup = arm_lookup, cache = 2000)
agents <-
list(
Agent$new(YahooLinUCBDisjointPolicy$new(0.2), bandit, name = "LinUCB"),
Agent$new(YahooEpsilonGreedyPolicy$new(0.2), bandit, name = "EGreedy"),
Agent$new(YahooRandomPolicy$new(), bandit, name = "Random")
)
Disjoint LinUCB vs EpsilonGreedy vs Random on Yahoo Data
simulations <- 1
horizon <- 10000000
bandit <- YahooBandit$new(k = 217L, d = 6L, arm_lookup = arm_lookup, cache = 2000)
agents <-
list(
Agent$new(YahooLinUCBDisjointPolicy$new(0.2), bandit, name = "LinUCB"),
Agent$new(YahooEpsilonGreedyPolicy$new(0.2), bandit, name = "EGreedy"),
Agent$new(YahooRandomPolicy$new(), bandit, name = "Random")
)
