Liking cljdoc? Tell your friends :D

org.replikativ.marketing.inference

NUTS inference returning portable chains, and convergence diagnostics.

fit samples a built model with Foerster's NUTS and returns a plain-data posterior (see io for persistence); diagnostics-summary checks it.

NUTS inference returning portable chains, and convergence diagnostics.

`fit` samples a built model with Foerster's NUTS and returns a plain-data
posterior (see `io` for persistence); `diagnostics-summary` checks it.
raw docstring

bp__org.replikativ.foerster.core_SLASH_deterministicclj

source

bp__org.replikativ.foerster.core_SLASH_factorclj

source

bp__org.replikativ.foerster.core_SLASH_observeclj

source

bp__org.replikativ.foerster.core_SLASH_sampleclj

source

bp__org.replikativ.foerster.effects_SLASH_chooseclj

source

bp__org.replikativ.foerster.effects_SLASH_deterministicclj

source

bp__org.replikativ.foerster.effects_SLASH_factorclj

source

bp__org.replikativ.foerster.effects_SLASH_observeclj

source

bp__org.replikativ.foerster.effects_SLASH_sampleclj

source

bp__org.replikativ.spindel.core_SLASH_awaitclj

source

bp__org.replikativ.spindel.core_SLASH_trackclj

source

bp__org.replikativ.spindel.effects.await_SLASH_awaitclj

source

bp__org.replikativ.spindel.effects.await_SLASH_await_finalizationclj

source

bp__org.replikativ.spindel.effects.savepoint_SLASH_savepointclj

source

bp__org.replikativ.spindel.effects.track_SLASH_trackclj

source

default-gatesclj

Convergence gates used by diagnostics-summary: R-hat strictly below 1.01, bulk and tail ESS of at least 100 for every parameter, no divergences.

Convergence gates used by `diagnostics-summary`: R-hat strictly below
1.01, bulk and tail ESS of at least 100 for every parameter, no divergences.
sourceraw docstring

diagnostics-summaryclj

(diagnostics-summary posterior)
(diagnostics-summary posterior gates)

Summarize convergence of a posterior from fit (or io/read!).

Returns

  • :parameters, a map from parameter name (e.g. [:beta "search"]) to {:mean :sd :rhat :ess-bulk :ess-tail :mcse :quantiles}, in draw order;
  • :max-rhat, :min-ess-bulk and :min-ess-tail over all parameters (nil when any of them is not finite) and :finite?;
  • :divergences and :depth-limit-visits from the sampler statistics;
  • :passes? when every parameter and the divergence count meet gates (default default-gates), and the :gates that were applied.
Summarize convergence of a posterior from `fit` (or `io/read!`).

Returns
- :parameters, a map from parameter name (e.g. [:beta "search"]) to
  {:mean :sd :rhat :ess-bulk :ess-tail :mcse :quantiles}, in draw order;
- :max-rhat, :min-ess-bulk and :min-ess-tail over all parameters (nil when
  any of them is not finite) and :finite?;
- :divergences and :depth-limit-visits from the sampler statistics;
- :passes? when every parameter and the divergence count meet `gates`
  (default `default-gates`), and the :gates that were applied.
sourceraw docstring

fitclj

(fit model)
(fit model
     {:keys [draws tune chains seed target-accept max-depth timeout-ms init
             gradient-backend metric jitter]
      :as options
      :or {max-depth 8
           target-accept 0.9
           gradient-backend :aot
           metric :variance
           timeout-ms 7200000
           draws 1000
           jitter false
           tune 1000
           chains 4
           seed 42}})

Fit a complete density block and detach its result from execution worlds. :draws excludes :tune. :timeout-ms bounds the whole fit, including gradient compilation (default 2 hours; nil disables the bound); on timeout the fit's world is cancelled and an exception is thrown. :gradient-backend defaults to :aot; :runtime selects raster runtime AD. :metric is :variance (default) or :fisher. :jitter defaults to false to keep the supplied initial position; true enables Foerster's per-chain jitter. Sampler diagnostics remain aligned with their retained chain and draw.

Fit a complete density block and detach its result from execution worlds.
:draws excludes :tune. :timeout-ms bounds the whole fit, including gradient
compilation (default 2 hours; nil disables the bound); on timeout the fit's
world is cancelled and an exception is thrown.
:gradient-backend defaults to :aot; :runtime selects raster runtime AD.
:metric is :variance (default) or :fisher. :jitter defaults to false to keep
the supplied initial position; true enables Foerster's per-chain jitter.
Sampler diagnostics remain aligned with their retained chain and draw.
sourceraw docstring

parameter-namesclj

(parameter-names model)

Parameter names of a built model, in posterior draw order, e.g. [:alpha "search"].

Parameter names of a built model, in posterior draw order, e.g. [:alpha "search"].
sourceraw docstring

cljdoc builds & hosts documentation for Clojure/Script libraries

Keyboard shortcuts
Ctrl+kJump to recent docs
←Move to previous article
→Move to next article
Ctrl+/Jump to the search field
× close