Repository navigation
Feat/contract api spine - #11
Closed
JavierGOrdonnez wants to merge 21 commits into
Closed
JavierGOrdonnez wants to merge 21 commits into
JavierGOrdonnez wants to merge 21 commits into
Conversation
added 9 commits
August 18, 2026 17:09
Consumers pass a table of samples plus configuration and receive typed results
in their own units. Training-file layout, Dakota configuration, preprocessing,
internal variable renaming, run directories and inverse transforms move behind
the boundary and are no longer reachable from a caller.
New package `itis_sumo.api`:
- errors.py SumoError taxonomy (input / result / engine), so consumers stop
classifying failures by matching on message text
- types.py PreprocessingSpec + VariableSpec in domain vocabulary only, and
frozen result dataclasses in original units, asdict-serializable
- _session.py internal fit-then-query engine owning validation, preprocessing
and run-directory lifetime: discarded on success, preserved on
failure with the path and stderr tail attached to the error
- workflows.py cross_validate() and evaluate_along_axes()
Also lands the grill outcome in SPEC: a VOCAB section (sample / variable =
parameter / response = quantity of interest), the consumer-contract goal line,
the constraints it implies with three unknowns parked rather than guessed, and
invariants V19cn-V27fq.
Supporting changes:
- export export_sumo_model/import_sumo_model and the E1 model store from the
public package surface; they were reachable only via funs_evaluate, which
V16qf forbids
- thread explicit seed and has_eval_id_column through the two wrapped
evaluation functions, so the facade never relies on the filename-substring
heuristic that caused B2
- docs: Glossary and Consumer API reference pages
- publish.yml accepts PEP 440 prerelease tags
- PORTING.md no longer claims the E1 work is unmerged
Backprops B10cs: a partial `at=` mapping raised a KeyError from
create_samples_along_axes, which requires a complete set of held values. The
API layer now materialises the default rather than assuming an internal helper
does.
302 tests pass, including 10 real unmocked Dakota runs through the new surface.
Completes the itis_sumo.api facade for all 8 workflows the mmux-vite Flask blueprint currently duplicates internally (blueprints/dakota.py), so a real port can proceed without discovering missing surface mid-cutover. - evaluate_uncertainty(): draws explicit per-variable uncertainty samples, evaluates the surrogate once, injects its own predictive std via the erfinv trick, and summarizes n_histograms realizations as a histogram + boxplot. This logic previously existed only inline in mmux-vite's Flask route and as a private test helper (_manual_uq_propagate); it is now a real itis-sumo library primitive (propagate_manual_uq_with_uncertainty + summarize_uncertainty_samples in evaluate/funs_evaluate.py). - optimize(): finds the Pareto-optimal trade-off front across one or more objectives. MOGA only ever supported a uniform search bound, not real uncertainty (get_bounds_uniform_distribution rejects anything else), so this introduces DomainSpec -- "where exploration is allowed" -- as a type distinct from DistributionSpec -- "what shape real-world uncertainty has" -- resolving the domain/distribution split that had been parked as an unknown pending real consumer evidence. - SumoSession is single-response by construction; MOGA is inherently multi-objective, so it gets its own fit-then-query pair (optimize_pareto_front) rather than forcing that mismatch into the session. _validate_samples is generalized to a list of responses so both paths share the same validation. Supporting: extracted _stderr_tail() to module level so both the session and the standalone MOGA path report failing run directories the same way.
propagate_manual_uq_with_uncertainty and evaluate_sobol_indices both sanitized input_vars/response_var/distributions at entry, then used those (now-sanitized) names to index preprocessor.input_variables / preprocessor.output_variables and to build DataFrame columns that preprocessor.transform() looks up by column name. The preprocessor is always fit on the caller's original names, so any variable name containing characters sanitize_varnames rewrites (spaces, parens, ...) caused a silent column-transform miss followed by a KeyError on the mapped_name lookup. create_manual_uq_samples had the same bug independently (it re-sanitized its own input_vars/distributions args). Dakota-safe names are already obtained correctly further down via preprocessor.input_variables[var].mapped_name, so the early sanitize_varnames calls were both redundant and actively harmful. Removed them; kept an explicit list() coercion in evaluate_sobol_indices where the removed sanitize call had incidentally normalized a tuple input to a list (pandas fancy-indexing needs a list, not a tuple). Surfaced by porting itis-sumo's Sobol route in mmux-vite, whose regression test exercises a real multi-word variable name end to end for the first time.
Adds generate_lhs_samples and generate_grid_samples to itis_sumo.api, wrapping the existing internal LHS design and full-factorial grid generation so consumers can draw input samples without touching itis-sumo internals. Needed so mmux_vite's sampling blueprint can stop depending on the vendored dakota subpackage for sample generation.
…ntegration helpers Now that mmux_vite's flaskapi fully consumes itis_sumo.api instead of these job-shaped web models, they have no remaining callers inside itis-sumo (only required_completed_jobs is used, by _session.py) and no remaining consumers outside it. Removes: - FunctionJob/JobVariableSelection pydantic models (preprocess/models.py) - preprocess/data_preprocessor_integration.py (only caller was the removed JobVariableSelection-based training-file helper; its other helpers had no callers anywhere in itis-sumo) - tests/test_data_preprocessor_integration.py Updates docs/reference/preprocess.md to drop the now-stale description.
added 8 commits
August 19, 2026 13:36
Collaborator
Author
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.
No description provided.