Forecasting with NumPyro - Feedback fehiepsi - #210
Merged
juanitorduz merged 1 commit intoJul 24, 2026
Conversation
juanitorduz
added a commit
that referenced
this pull request
Aug 4, 2026
… – Open‑Source Edition] (#202) * init init2 revision initl ijof labs address rm legacy rm * feedback1 init (#204) * IJF paper improvements: trim, cuthbert, referee-proofing (#205) * IJF paper improvements: trim, cuthbert, referee-proofing Tighten the manuscript toward the ~20-page special-section brief (23 pages including references), add cuthbert and scan-vs-Kalman flexibility in the ecosystem section, and address IJF-audience referee points: reconciliation terminology, proper scoring rules, vmap backtesting, related-substrate positioning, and softened TSB claim. Rebuild PDF with verified figures. Co-authored-by: Cursor <[email protected]> * Revise availability-TSB and censored-demand case studies Add modified TSB equations and scan-flexibility argument to Section 4.2. Rewrite Section 4.3 around the demand blog: remove mismatched censoring_12_0.png, expand censored-likelihood narrative, and focus Figure 3 on demand_25_0.png with an accurate caption. Add amsmath for equation blocks and cross-links from Section 3 and the community layer. Co-authored-by: Cursor <[email protected]> * rerun * add code back * improve image --------- Co-authored-by: Cursor <[email protected]> * Add structured-covariate section and pyrenew reference (#206) Promote structured covariate handling to a dedicated Section (sec:covariates) with the event_basis listing relocated from the HSGP case study, expand the effect-handler paragraph, thread the affordance through the abstract, intro, and conclusion, and add the pyrenew reference. Rebased onto the merged IJF improvements from PR #205. Co-authored-by: Cursor <[email protected]> * references (#207) * Address Theo's final review (PR #202 round 2) (#208) Apply all 18 inline comments from review 4583325705: reword the joint-draws/election passage, drop Optax from the primitives sentence, clarify Flax's role, state the scan-vs-Kalman trade-off where scan's ease is claimed, motivate BlackJAX/flowMC via difficult posterior geometries, replace the stale dynamax cite with dynestyx (which specifies SSMs as NumPyro models and calls cuthbert), hyphenate end-to-end, set off the cross-validation aside with dashes, rename affordance to capability throughout, use standard Normal(0,1)/ HalfNormal(1) hyperpriors in the event-basis listing, explain why non-centered reparameterization helps NUTS, spell out TSB at first use, reword "forking the estimator", add the auxiliary-noise-scale weighting sentence to the calibration section, mention pretrained models as boosting layers, and cite the Chronos-2, Toto, and TimesFM papers instead of checkpoint/port repos. Co-authored-by: Claude Fable 5 <[email protected]> * Expand numpyro_forecast description and fresh-retail stockout example (#209) * Expand numpyro_forecast description and censored-demand case study - Section 2: describe numpyro_forecast's functional and object-oriented APIs, rolling/expanding-window backtesting (time-sliced cross-validation, including the vectorized vmap variant), and metrics such as the continuous ranked probability score and empirical interval coverage. - Section 5.3: extend the censored-demand case study with the multiplicative availability-factor approach on FreshRetailNet-50K, the full-availability counterfactual, and a new figure (series 22::267). - Report the 50K-series scaling data point (approximately ten minutes on a single GPU via Modal) and the chunked host-offload posterior predictive sampling. - Section 3.4: change "new benchmarks" to "systematic benchmarks" for consistency with the new wall-clock data point. - Add references: FreshRetailNet-50K dataset paper (Wang et al., 2025) and the "Forecasting Retail Demand Under Stockouts" worked example. Co-Authored-By: Claude Fable 5 <[email protected]> * Review pass: correctness, bibliography, style, and layout fixes Correctness: numpyro.set_platform("cuda") (the "gpu" string is invalid in NumPyro 0.21), accurate handlers.mask description (likelihood masking, not AR-state freezing), accurate optimizer claims (numpyro.optim exists; Optax accepted directly), obermeyer2022pyro cited for epidemiological (not compartmental) models, export-path argument names (predictive_batch_size/predictive_device), Listing 2 caption notes elided conditioning and horizon guards. Bibliography: riutort2022practical is a 2023 Statistics and Computing journal article (was 2022 @inproceedings), gorinova2019automatic is ICML 2020 pp. 3648-3657 (was 2019 with wrong pages), six @software entries converted to @misc with howpublished URLs (silences all bibtex warnings). Style: Acknowledgments (American spelling), abstract trimmed 159 to 149 words per the IJF 100-150 limit (also drops code notation), time-series cross-validation naming aligned, pre-trained/dataset hyphenation unified, small grammar fixes, SVI/NUTS acronyms replaced with full terms at first package mention. Layout: float parameter tuning and top-aligned float pages, both code listings are now unbreakable floats (fixes Listing 2 splitting across pages), weak [h] figure placements normalized to [!htbp], availability figure slightly reduced so text flows beneath it. No more lone vertically centered figure pages; build is clean (0 undefined refs, 0 overfull boxes, 0 bibtex warnings); still 29 review-mode pages. Co-Authored-By: Claude Fable 5 <[email protected]> * Add IJF submission checklist Captures the special-section logistics (dates, article type, cover letter, guest editors), the double-anonymization work the IJF guide requires before submission, the declarations to prepare, and referee-risk notes from the call's evaluation criteria. Co-Authored-By: Claude Fable 5 <[email protected]> --------- Co-authored-by: Claude Fable 5 <[email protected]> * corrections init (#210) * rm email --------- Co-authored-by: Cursor <[email protected]> Co-authored-by: Claude Fable 5 <[email protected]>
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.
Review from fehiepsi :)