Conversation
Translates R's adam-ces.R to Python with full CES pipeline: creator, filler, initialiser, cost function, and CES/AutoCES model classes. Supports all four seasonality modes (none, simple, partial, full). 188 parity tests pass against R reference outputs. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add six standalone simulators (sim_es, sim_ssarima, sim_sma, sim_oes, sim_ces, sim_gum) and ADAM.simulate() — all driving the shared C++ adamCore::simulate kernel. Each accepts an R-name string or a Python callable for ``randomizer``, enabling plug-in-numbers R-parity tests that match R to floating-point via the same kernel. SimulateResult dataclass mirrors R's smooth.sim / oes.sim S3 lists with __repr__ matching print.smooth.sim / print.oes.sim. Version bumped to 1.0.6. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Syncing the branches
Extract simulateADAMCore() helper out of simulate.adam (matrix prep, distribution-aware error sampling, occurrence mask, C++ call). The public simulate.adam now wraps that helper with seed + ts/zoo timing + adam.sim assembly, behaviour unchanged. The new simulate.om calls the same helper, applies omLinkFunction to map the latent state-space output to a probability series, and draws rbinom for 0/1 occurrence indicators. simulate.omg recursively calls simulate.om on the two sub-models and combines via omgLinkFunction on the latent series. Two side-effect adjustments inside simulateADAMCore so the helper covers om/omg: handle is.character(object$occurrence) (om stores a scheme string, not a fitted occurrence object) and reconstruct the state cube's lag head from profileInitial when nrow(states) == obsInSample (om's compact state shape). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
OM.simulate overrides ADAM.simulate: calls super() to get the latent ETS draw, applies om_link_function to convert to probability, draws rng.binomial(1, prob) for the 0/1 indicator series. OMG.simulate calls model_a.simulate / model_b.simulate with derived seeds and combines the two latent series via omg_link_function. SimulateResult gains a ``latent`` field that exposes the pre-link state-space output for OMG to consume. The default randomizer for OM is a Gaussian sampler keyed on the empirical residual std — substitutes for the missing ``plogis`` branch in generate_errors(). Callable randomizers bypass that and enable the plug-in-numbers R-parity trick. ADAM.simulate tweaked to accept the structured-dict ``persistence`` form OM stores (falls back to ``vec_g``). R-side simulateADAMCore gained a one-block randomizer override so the same plug-in-numbers trick works on the R side too. 10 new smoke tests + 4 R-parity tests (1 passing, 3 xfailed with explanatory notes on the OM-fit-parity / Python-numerics / R-side-ARMA-bug they each expose). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Closes the historical ULP gap on ADAM initial="optimal" / "two-stage" vcov by routing both languages' msdecompose() through the same Armadillo pivoted-QR with a scale-invariant LINPACK-dqrls-style rank cutoff. R-parity bench: max |ΔSE/SE| drops from ~2% to 1e-8 on ETS and 1e-4 on ARIMA (the FD-Hessian floor). The vcov R-parity test tightens from rtol=2e-2 to rtol=1e-3. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The historical 5% tolerance reflected the OLS-ULP-into-x0-into-NLopt cascade that pushed Python's optimiser onto a slightly different B than R's. With (a) the per-parameter relative FD Hessian step in hessianCore.h and (b) the shared olsCore.h backend for msdecompose, that cascade is gone — R and Python converge to the same B on the ANN/AAN/ARIMA/OM-MNN scenarios, the analytical formula is a closed form, and multicov agrees to machine precision (worst case ~1e-14, mostly at or below 1e-15). Empirical multicov is also at machine precision because the underlying ferrors C++ backend is shared. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Resolve conflicts: - src/headers/lowess.h: delete (moved to greybox on origin/Python) - python/pyproject.toml: take origin's nlopt (unpinned) and scipy<1.19.0 - python/src/smooth/__init__.py: union CES/AutoCES with AutoMSARIMA/AutoOM - python/src/smooth/adam_general/core/__init__.py: union CES/AutoCES with AutoMSARIMA
Bringing up to speeeeeeeeeeed
- ces_model.py: annotate a/b dicts, B and class attributes, cast Literal seasonality, replace dict.get key with explicit lambda - utils.py: type-ignore on the _ols pybind import (mirrors other C++ extension imports) - test_ces.py: add pytestmark = pytest.mark.r_parity so the strict 1e-9-tolerance R-parity tests opt out of the default run, matching test_msdecompose_r_parity.py / test_multicov_r_parity.py / etc.
NEWS.md gets a v1.0.6 entry for the new CES + AutoCES classes (port of R's ces() / auto.ces()). README.md lists CES in the model overview, the per-model wiki links, and adds a usage block with both fixed-seasonality CES() and AutoCES seasonality selection.
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
2 participants
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.
What issue does this PR fix?
Closes #350
Closes #349
Language