anglican.bamc — anglican 1.1.0

Bayesian ascent Monte Carlo Options: :predict-candidates (false by default) - output all samples rather than just those with increasing log-weight

->entry clj/s≠

(->entry bandit-id value past-reward)

clj

Positional factory function for class anglican.bamc.entry.

cljs

Positional factory function for anglican.bamc/entry.

->multiarmed-bandit clj/s≠

(->multiarmed-bandit arms new-arm-belief new-arm-count new-arm-drawn)

clj

Positional factory function for class anglican.bamc.multiarmed-bandit.

cljs

Positional factory function for anglican.bamc/multiarmed-bandit.

add-bandit-predict clj/s

(add-bandit-predict state)

add bandit arms and counts as a predict

add-trace-predict clj/s

(add-trace-predict state)

adds trace as a predict

backpropagate clj/s

(backpropagate state)

back propagate reward to bandits

bandit-id clj/s

(bandit-id smp state)

bayesian-belief clj/s protocol

Bayesian belief

bb-as-prior clj/s

(bb-as-prior belief)

returns a belief for use as a prior belief

bb-sample clj/s

(bb-sample belief)

returns a random sample from the belief distribution

bb-sample-mean clj/s

(bb-sample-mean belief)

returns a random sample from the mean belief distribution

bb-update clj/s

(bb-update belief evidence)

updates belief based on the evidence

entry cljs

fresh-bandit clj/s

bandit with no arm pulls

initial-reward-belief clj/s

uninformative mean reward belief

initial-state clj/s

initial state for MAP estimation

map->entry clj/s≠

clj

(map->entry m__7585__auto__)

Factory function for class anglican.bamc.entry, taking a map of keywords to field values.

cljs

(map->entry G__23576)

Factory function for anglican.bamc/entry, taking a map of keywords to field values.

map->multiarmed-bandit clj/s≠

clj

(map->multiarmed-bandit m__7585__auto__)

Factory function for class anglican.bamc.multiarmed-bandit, taking a map of keywords to field values.

cljs

(map->multiarmed-bandit G__23514)

Factory function for anglican.bamc/multiarmed-bandit, taking a map of keywords to field values.

multiarmed-bandit cljs

not-a-value? clj/s

(not-a-value? value)

true when new value must be sampled

record-choice clj/s

(record-choice state bandit-id value past-reward)

records random choice in the state

reward-belief clj/s

(reward-belief sum sum2 cnt)

returns reification of bayesian belief about the mean reward of an arm

select-value clj/s

(select-value bandit log-p)

selects value corresponding to the best arm

update-bandit clj/s

(update-bandit bandit value reward)

updates bandit's belief