anglican.almh — anglican 1.0.0

Adaptive LMH

->predict clj

(->predict value choices)

Positional factory function for class anglican.almh.predict.

add-choice-predicts clj

(add-choice-predicts state)

adds predicts for choice statistics to the state

add-pending-choice clj

(add-pending-choice predict choice-id)

adds a pending choice to the predict

award clj

(award state entry)

distributes rewards between random choices; returns updated state

combined-predicts clj

(combined-predicts state)

combines current predicts with last predicts for which no current predict is available, assigning the latters +not-a-predict+ as the value; returns combined predicts

initial-state clj

initial state for ALMH

initialize-last-predicts clj

(initialize-last-predicts state)

initializes last predicts from current predicts, called on the first end state; returns updated state

log-entry-probability clj

(log-entry-probability state entry)

computes log probability of the entry given state

map->predict clj

(map->predict m__7585__auto__)

Factory function for class anglican.almh.predict, taking a map of keywords to field values.

next-state clj

(next-state state entry)

produces next state given current state and the trace entry to resample

prev-state clj

(prev-state state next-state entry)

produces previous state given the current and the next state and the resampled entry by re-attaching new rdb to the original state

select-entry clj

(select-entry state)

selects trace entry based on reward beliefs

state-update clj

(state-update state)

computes state update --- fields that must be transferred to the state before running random choice continuation

trace-weights clj

(trace-weights state)

constructs a vector of trace weights

ucb clj

(ucb total-count)

returns function computing UCB of average reward

update-choice-count clj

(update-choice-count state entry)

updates total count for the chosen entry

update-reward clj

(update-reward [sum cnt] reward discnt)

updates choice reward with new evidence

update-rewards clj

(update-rewards choice-rewards pending-choices reward discnt)

updates rewards in pending choices; returns updated choice-rewards

utility clj

(utility state entry)

computes state utility, used to determine the acceptance log-probability as (next-utility - prev-utility)