0.2.5

Fix

• fix rewardplot (#123)

0.2.4

Change

• move reward from History to Recommendation (#121)

0.2.3

New

• BaseBandit support update_action and remove_action methods (#120)

Change

• recommendation class (#117)

Fix

• fix and unify the return value of BaseBandit.get_action() when the action storage is empty (#118)

0.2.2

Fix

• unify the data usage in all simulations, fixed #95 (#107)
• multi-action history, fixed #58 (#110)
• fix rewardplot.calculate_cum_reward (#112)
• remove query_vector in the Exp3 model and use the history instead, fixed #108 (#112)

Change

• History and HistoryStorage interfaces are changed (#110)

0.2.1

Fix

• fix striatum.utils import error (#105)

0.2.0

Change

• ActionStorage (#99)
• reorder bandit __init__ parameters and change some default values (#102)
• default n_actions in BaseBandit.get_action is chaged to None (#103)
• BaseBandit.get_action add support for n_actions=None and n_actions=-1 (#103)

Fix

• remove the generators in Exp3 and UCB1 (#101)

0.1.1

Fix

• fix exp3 random_state; fixed #86 (#91)
• correct setup configuration; fixed #87 (#88)

0.1.0

New

• add reward calculation and plot methods in each bandit policy (#38)
• add trivis ci using tox (#44)
• Python3 support (#44)
• generating documentation using sphinx and host it on readthedoc (#44)
• BaseBandit.get_action_with_id (#63)

Change

• refactor bandit algorithms to allow multiple actions and rewards (#37)
• use Action object instead of action_id during policy initialization (#37)
• use "expert advice probability vectors" instead of "scikit-learn models" as input for Exp4p (#37)
• better simulation coding style (#47, #51, #53, #56, #72, #73, #74, #75, #76, #78)
• better coding style (#50, #62, #66, #79)

Fix

• fix the parameter updating bugs (query_vector calculation) in Exp4p (#37)
• fix bugs in Exp4.P (#66)
• remove generator in LinUCB (#54)
• remove generator in Exp4.P (#67)
• remove generator in LinTompSamp (#80)

0.0.1

New

• implement LinUCB ( #10 )
• implement UCB1 ( #10 )
• implement EXP3 ( #10 )
• implement EXP4P ( #10 )
• implement Thompson Sampling for Contextual Bandits with Linear Payoffs ( #10 )
• provide unit test for each bandit algorithm ( #10 )
• simulation on fake data ( #10 )
• benchmark using movielens ( #29 )