Fit and check declared, causally sufficient DAGs from complete numeric rows. The default mechanisms are linear Gaussian with independent standardized noise. Fitting is not causal discovery, and diagnostics do not establish the graph or its counterfactual coupling. No external numerical runtime.
Fit and check declared, causally sufficient DAGs from complete numeric rows. The default mechanisms are linear Gaussian with independent standardized noise. Fitting is not causal discovery, and diagnostics do not establish the graph or its counterfactual coupling. No external numerical runtime.
(abduct fitted row)Reusable cf abduction for one COMPLETE observation at fixed fitted parameters. Continuous invertible noise is exact; custom discrete mechanisms sample conditional noise. No latent/parameter posterior is inferred. Use cf/query with evidence for incomplete observations instead.
Reusable cf abduction for one COMPLETE observation at fixed fitted parameters. Continuous invertible noise is exact; custom discrete mechanisms sample conditional noise. No latent/parameter posterior is inferred. Use cf/query with evidence for incomplete observations instead.
(bootstrap fitted training statistic)(bootstrap fitted
training
statistic
{:keys [seed replicates level] :or {replicates 200 level 0.95}})IID row bootstrap: refit ALL mechanisms and compute a finite scalar (statistic fitted-model). Reports percentile interval, point estimate, and every draw. Defaults: 200 replicates, 95% interval. Failures abort rather than silently dropping singular refits. This is sampling uncertainty conditional on graph/family, not uncertainty about causal assumptions. Fitting callbacks must be deterministic for seeded reproducibility.
IID row bootstrap: refit ALL mechanisms and compute a finite scalar (statistic fitted-model). Reports percentile interval, point estimate, and every draw. Defaults: 200 replicates, 95% interval. Failures abort rather than silently dropping singular refits. This is sampling uncertainty conditional on graph/family, not uncertainty about causal assumptions. Fitting callbacks must be deterministic for seeded reproducibility.
(distribution fitted node values)Conditional structural law at node. Only declared parents reach its callback.
Conditional structural law at `node`. Only declared parents reach its callback.
(evaluate fitted held-out)(evaluate fitted
held-out
{:keys [seed permutations alpha] :or {permutations 199 alpha 0.05}})Evaluate on separate held-out IID complete rows. Reports node log scores, scalar noise moments and, for built-in Gaussian fits, RMSE and 90% coverage. Seeded permutation tests use the maximum absolute correlation between centered/scaled x,x² and y,y²: residual-vs-parent and residual-vs-residual. Holm correction controls family-wise error at :alpha (default .05). Defaults: 199 permutations. This test detects some nonlinear dependence, not all; no rejection is not evidence of causal correctness or Gaussianity. Report :rejected-checks, never a 'causally valid' flag.
Evaluate on separate held-out IID complete rows. Reports node log scores, scalar noise moments and, for built-in Gaussian fits, RMSE and 90% coverage. Seeded permutation tests use the maximum absolute correlation between centered/scaled x,x² and y,y²: residual-vs-parent and residual-vs-residual. Holm correction controls family-wise error at :alpha (default .05). Defaults: 199 permutations. This test detects some nonlinear dependence, not all; no rejection is not evidence of causal correctness or Gaussianity. Report :rejected-checks, never a 'causally valid' flag.
(fit g rows)(fit g rows {:keys [fitters] :as options})Fit a declared DAG, defaulting to linear Gaussian mechanisms. rows are
IID observational records, with all graph nodes observed. Hidden common
causes (:bidirected edges) are refused. :fitters maps node -> custom fitter
with the linear-gaussian signature, returning {:distribution (fn [parents]
law)}. That law must implement Distribution and PMechanism. A custom
:noise-score maps noise to a finite scalar for diagnostics (default identity).
Independent noise and the graph's causal meaning are author assumptions.
Fit a declared DAG, defaulting to linear Gaussian mechanisms. `rows` are
IID observational records, with all graph nodes observed. Hidden common
causes (:bidirected edges) are refused. :fitters maps node -> custom fitter
with the `linear-gaussian` signature, returning {:distribution (fn [parents]
law)}. That law must implement Distribution and PMechanism. A custom
:noise-score maps noise to a finite scalar for diagnostics (default identity).
Independent noise and the graph's causal meaning are author assumptions.(linear-gaussian node parents rows)Fitter contract: (fitter node ordered-parents training-rows) -> mechanism. OLS mean, maximum-likelihood residual scale; refuses insufficient rows, constant/collinear parents and effectively deterministic residuals.
Fitter contract: (fitter node ordered-parents training-rows) -> mechanism. OLS mean, maximum-likelihood residual scale; refuses insufficient rows, constant/collinear parents and effectively deterministic residuals.
(model fitted)An ordinary foerster model spin: one sample site per node, node IDs as addresses. Use with inference, cf/query, attribution and trace policies.
An ordinary foerster model spin: one sample site per node, node IDs as addresses. Use with inference, cf/query, attribution and trace policies.
(replay fitted noise)(replay fitted noise interventions)Synchronous structural evaluation at supplied noise. interventions maps
nodes to hard values (not trace policy maps). For richer interventions and
compositional worlds use model with the existing cf APIs.
Synchronous structural evaluation at supplied noise. `interventions` maps nodes to hard values (not trace policy maps). For richer interventions and compositional worlds use `model` with the existing cf APIs.
(simulate fitted n)(simulate fitted n {:keys [seed interventions]})IID draws from the fitted model, optionally under hard interventions. :seed gives a reproducible stream without changing the process RNG.
IID draws from the fitted model, optionally under hard interventions. :seed gives a reproducible stream without changing the process RNG.
cljdoc builds & hosts documentation for Clojure/Script libraries
| Ctrl+k | Jump to recent docs |
| ← | Move to previous article |
| → | Move to next article |
| Ctrl+/ | Jump to the search field |