Interactive graphical models: model() worlds with live, editable beliefs#20
Open
koaning wants to merge 8 commits into
Open
Interactive graphical models: model() worlds with live, editable beliefs#20koaning wants to merge 8 commits into
koaning wants to merge 8 commits into
Conversation
…h demo Bars signal mode (black static, blue editing); reset is the single way out of edit mode. Model variables display as live wired charts (marimo prefers _display_), a bare World renders its graph, and learn.py gains a live factor-graph + joint-atoms prototype. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Observable Plot sorts ordinal domains alphabetically by default, so the CPT heatmaps and belief slots disagreed with the variable charts' outcome order. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…uery The joint is now a product of factors (priors + families); queries run min-fill variable elimination, beliefs fit via IPF over adjustment factors, and whole-graph repaints use one two-pass sum-product calibration on forest factor graphs (VE fallback when loopy). Cost follows treewidth, not variable count: a 200-test star repaint drops 12.3s -> 0.025s, and models the old enumerator could not represent (2^201 rows) answer instantly and exactly. Edge messages are stored for the upcoming message-passing visualization. The enumerator survives as joint(), cross-checked by randomized oracle tests. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The factor view renders the real engine factors (prior + CPT heatmaps) between the variables they touch, with a dashed belief slot per variable that lights up when a belief is asserted. messages=True adds a two-row strip per edge showing the calibration's actual sum-product messages, repainted on every belief change. Replaces the hand-built learn.py prototype with the one-liner. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Each verified three ways: an independent in-test brute-force enumeration sharing the model's CPT constants, published literature values (Wikipedia sprinkler 0.3577/0.6467, R&N alarm 0.284, Monty (n-1)/(n(n-2)) for 3-5 doors), and cross-engine paths (hard given vs point-mass believe, Jeffrey mixture identity for soft beliefs, exact Fraction arithmetic end to end). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a probabilistic-graphical-model layer to dicekit: inside
with model() as m:a bareDiceis a root variable,Dice.from_function(fn, parents=[...])adds a conditional dependent, and operator expressions become derived variables over one exact joint, queryable viap()/m.query()with hard (given=) and soft (Jeffrey) evidence. Beliefs are now stateful and update in place:dice.believe()/forget()condition every subsequent query, multiple beliefs are satisfied jointly via iterative proportional fitting (order-independent), and lazy marginal caches invalidate automatically.m.show()renders the model as a DAG of live charts where every node has an in-widget edit toggle — drag bars to assert a belief and all other nodes repaint through kernel-side widget wiring (no notebook re-runs, no editor-cell boilerplate); displaying a bare variable or world gives the same live view. Also fixes a KeyError for derived parents infrom_function, and the learn notebook gains a DAG prototype plus a live factor-graph and joint-atoms visualization. Includes 30+ new tests (67 passing) and notebook-driven builds viamake build.🤖 Generated with Claude Code