Open
Conversation
Member
isaacsas
commented
Mar 23, 2026
isaacsas
commented
- Needs the Project.toml, and CI workflows updated to remove sources and references to this branch.
- Blocked waiting on a SciMLBase release with Ensemble rng redesign SciMLBase.jl#1252
Make the integrator the single source of truth for RNG (Phase 1 Step 4). Remove rng field from all aggregator structs, JumpProblem, and variable-rate callback structs. All code now reads get_rng(integrator) at runtime. Key changes: - Remove RNG type param from AbstractSSAJumpAggregator and all concrete aggregators - Remove rng from build_jump_aggregation and all aggregate() signatures - Add rng field to SSAIntegrator with has_rng/get_rng/set_rng! interface - Add resolve_rng utility for rng/seed/fallback priority resolution - Remove rng from VR_FRMEventCallback and VR_DirectEventCache - JumpProblem rng kwarg now forwards to solver via solkwargs - Simplify solve.jl: remove _derive_jump_seed, simplify __jump_init - Update SimpleTauLeaping/SimpleExplicitTauLeaping to use rng kwarg - Add rng_kwarg_tests.jl covering all three solve pathways - Update docs with RNG control section and FAQ entry - Bump version to 9.23.0, SciMLBase compat to 2.144 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Breaking: JumpProblem no longer accepts an `rng` keyword argument. Passing `rng` now throws an ArgumentError directing users to pass it to `solve` or `init` instead. This fixes inconsistent RNG priority ordering across solver pathways (SSAStepper, ODE/SDE, tau-leaping) and makes the integrator the single source of truth for RNG state. - Simplify resolve_rng to 2-arg form (remove fallback_rng) - Remove jump_prob.kwargs rng fallback from tau-leaping solvers - Update all tests to pass rng to solve/init instead of JumpProblem - Update docs and HISTORY.md with breaking change notes Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
…ecking - Revert <:Any back to named type params in ssajump.jl (minimize diff from master) - Revert solkwargs to simpler ternary form in problem.jl (rng removal made filter unnecessary) - Add type checking to SSAIntegrator set_rng! to match StochasticDiffEq pattern Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The GPU tau-leaping kernel uses PassthroughRNG internally; the StableRNG was never consumed by the EnsembleGPUKernel path. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
solve(; rng = nothing) falls through to default_rng() via resolve_rng, so the ternary guards are redundant. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Comparing continuous-valued jump times is more robust than discrete final state values, which could collide by chance. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Check jump_u thresholds directly via init, and verify both event times and post-event thresholds from ensemble solve. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Resolves conflicts by combining integrator RNG (get_rng) with u_modified!(true) fix, and using robust first_jump_time helpers in tests that account for the initialization save. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Remove StochasticDiffEq from [deps] (test-only, belongs in [extras]) - Move EnsembleThreads data race tests from ensemble_problems.jl to thread_safety.jl with stronger assertions (400 trajectories, allunique on first reaction times, all 3 VR aggregators, adaptive SDE solver) - Renumber remaining ensemble_problems.jl sections Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- ensemble_problems.jl: check initial jump_u threshold (u[2]) instead of post-event (u[3]) now that u_modified!(true) produces init save - extended_jump_array_remake.jl: rework to solve → remake → init → compare jump_u, verifying fresh thresholds after remake - geneexpr_test.jl: remove unnecessary isnothing(rng) branch - linearreaction_test.jl: make rng const for type stability - variable_rate.jl: switch getmean from seed integer to StableRNG - rng_kwarg_tests.jl: add SDE+VR different seeds test, SDE+VR no-rng functional test, seed kwarg reproducibility and different-seeds tests for all pathways (SSAStepper, ODE+VR, SDE+VR), use first_jump_time helper for robust event time extraction Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
… strengthen thread_safety assertions - docs/advanced_point_process.md: move rng from JumpProblem to solve - rng_kwarg_tests.jl: add seed kwarg tests for SimpleTauLeaping (tests 13-14), renumber 15-20 - thread_safety.jl: SSAStepper EnsembleThreads now runs 400 trajectories and asserts allunique first jump times Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Create ThreadSafety.yml workflow: runs thread_safety.jl with --threads=4 - Add ThreadSafety group to Tests.yml matrix (serial) and runtests.jl - Move thread_safety.jl from InterfaceII to ThreadSafety group - Fix shared rng data race: SSAStepper ensemble tests now use task-local default_rng() instead of a shared StableRNG across threads Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
…-scaling When MTKBase/Catalyst constructs MassActionJumps, the symbolic rate expressions already include stoichiometric scaling (e.g. k/3! for 3X → Y), so they pass scale_rates=false. However, update_parameters! defaulted to scale_rates=true, causing every subsequent parameter update path (reset_aggregated_jumps!, remake, finalize_parameters_hook!) to double-scale the rates. The fix stores the intended scaling behavior on the struct as rescale_rates_on_update, which update_parameters! now reads as its default. This field is propagated through all merge/combine paths and validated for consistency when merging multiple MassActionJumps. Also fixes SpatialMassActionJump conversion constructor to default scale_rates=false since it receives already-scaled rates. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The mapper now conditionally applies scalerates! based on scale_rates, matching MTKBase JumpSysMajParamMapper behavior. Previously it ignored scale_rates entirely, so the test would pass even with the old buggy default of scale_rates=true. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Document that JumpProblem contains mutable state and is not thread-safe for concurrent solve calls. Note that EnsembleProblem handles isolation automatically via SciMLBase per-task deepcopy and per-trajectory RNG. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
…structs Move mutable rate working-copy (maj_rates) out of MassActionJump and into all 10 aggregation structs + RxRates. Rates are now lazily materialized via fill_scaled_rates! during initialize! instead of eagerly in the JumpProblem constructor. This fixes aliasing bugs where multiple JumpProblems sharing the same MAJ could silently corrupt each others rates. Key changes: - Add maj_rates field to all aggregation structs and RxRates - Add fill_scaled_rates! dispatcher for parameterized vs explicit MAJs - Update evalrxrate to 4-arg signature taking explicit maj_rates - Propagate maj_rates through calculate_jump_rate, rejectrx, bracketing - Remove update_parameters!, finalize_parameters_hook! - Simplify reset_aggregated_jumps! (remove update_jump_params kwarg) - JumpProblem constructor keeps original parameterized MAJ - Add merge guards for mapper-backed MAJs - Add SpatialMassActionJump guard for parameterized MAJs - Remove dead 1-arg mapper callables - Update all tests for lazy rate materialization pattern Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
use integrator rng API
- Narrow MAJ merge guards to only block custom (non-MassActionJumpParamMapper) mapper types. Built-in parameterized MAJ merges via varargs and JumpSet vectors work correctly and are tested in ssa_callback_test.jl. - Fix 3-arg evalrxrate call in test/spatial/reaction_rates.jl to use eval_massaction_rate which handles both MassActionJump and SpatialMassActionJump correctly. - Add fill_scaled_rates! call before spatial rate tests to populate maj_rates. - Update scale_rates_field_test.jl merge test to verify built-in merges succeed and custom mapper merges error. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- setup_majump_to_merge now copies arrays before creating the merge accumulator, preventing in-place mutation from corrupting the original MAJ net_stoch/reactant_stoch arrays - massaction_jump_combine(::Nothing, ::MassActionJump) now creates a safe copy via setup_majump_to_merge instead of returning the original MAJ - Remove scale_rates and useiszero as JumpProblem kwargs; they are now properties of the MassActionJump itself. Passing them raises an informative ArgumentError Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Merge master into revamp_maj_initialization, bringing in: - OrdinaryDiffEqCore package extension for __init ambiguity fix - ExtendedJumpArray lmul!/ldiv! bug fix - SDE+jump callback duplication test - Doc updates for Catalyst v16 Reverted master update_jump_params kwarg addition to SSAIntegrator reset_aggregated_jumps! and removed associated test, since this branch removes that kwarg entirely (throws ArgumentError if passed). Dropped OrdinaryDiffEqCore =3.15 pin in favor of master compat bound. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Move rng kwarg from JumpProblem to solve/init in merged test files (extended_jump_array.jl, callbacks.jl) - Remove stale seed! import from JumpProcesses.jl (QA failure) - Add fill_scaled_rates! in spatial bracketing test to populate maj_rates - Remove old prob_func(prob, i, repeat) signature in thread_safety.jl (use default EnsembleProblem which already deepcopies) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
… update bracketing test - Add fill_scaled_rates! no-op for SpatialMassActionJump (rates stored in struct, no working copy needed). Placed in spatial_massaction_jump.jl since the type is defined after jumps.jl in the include order. - Update test/bracketing.jl: pass maj_rates to get_majump_brackets, add maj_rates field to DummyAggregator, call fill_scaled_rates! to init rates. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Bug 1: A single parameterized MassActionJump with scalar param_idxs
(e.g. param_idxs = 1) would segfault when passed directly to JumpProblem
without merging, because the scalar MassActionJumpParamMapper{Int} had no
callable method. Fix: convert scalar param_idxs to a one-element vector at
construction time, and add a JumpSet constructor for single-reaction
parameterized MAJs to normalize them to collection format.
Bug 2: Merging two single-reaction parameterized MAJs with vector-backed
param_idxs would mutate the first MAJ's mapper via aliasing in
to_collection. Fix: to_collection now copies the param_idxs vector.
Also: add docstring for the mapper callable API, remove dead code for
Int-typed mappers, and add tests for both bugs.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
…TORY.md - Move mapper callable API docstring from the method to MassActionJumpParamMapper struct where it belongs - Add HISTORY.md entries for v10 breaking changes: scale_rates/useiszero removal from JumpProblem, immutable parameterized MAJ design, update_parameters! removal with reset_aggregated_jumps! guidance, custom mapper 3-arg callable requirement, scalar param_idxs normalization Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Revamp MassActionJump initialization
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.