Skip to content

chore(deps): update dependency madgik/exareme2 to v1.2.1 (federation override) - #49

Draft
renovate[bot] wants to merge 1 commit into
mainfrom
renovate/federations-madgik-exareme2-1.x
Draft

renovate[bot] wants to merge 1 commit into
mainfrom
renovate/federations-madgik-exareme2-1.x

Conversation

@renovate

@renovate renovate Bot commented Oct 1, 2026 •

Copy link
Copy Markdown
Contributor

This PR contains the following updates:

Package Update Change
madgik/exareme2 minor 1.1.0 → 1.2.1

Release Notes

madgik/exareme2

v1.2.1

Compare Source

Exaflow 1.2.1 Patch Release Notes

Linear Regression Cross-validation

  • Corrected metrics: Predictor counts now reflect encoded covariates. RMSE, MAE, R², and the F diagnostic are calculated independently, preserving valid metrics when another is undefined.
  • Undefined results: Metric summaries return [null, null] when any fold produces an undefined or infinite value.
  • Renamed RMSE field: Replaced mean_sq_error with root_mean_sq_error. API consumers must update the field name and handle null summaries.
  • Expanded validation: Added reference parity and edge-case tests, updated validation fixtures, and documented metric calculations and limitations.

Runtime & Diagnostics

  • Floating-point tolerance: Local UDF result comparisons now tolerate small rounding differences in non-whole float pairs, using relative tolerance 1e-9 and absolute tolerance 1e-12. Whole-valued floats and non-float values retain exact comparison.
  • Mismatch logging: Inconsistent local UDF responses now log the function name, result indices, and differing results before raising an error.

v1.2.0

Compare Source

Exaflow 1.2.0 Release Notes

New Algorithms & Statistical Methods

  • Quartiles: Added the quartiles analysis to estimate Q1, median, and Q3
    using iterative federated histogram refinement. Bin count, iteration limit,
    and stopping threshold are configurable; results include the achieved
    cumulative fraction when it differs from the requested quartile.
  • Binned Mann-Whitney U test: Added binned_mann_whitney_u_test for comparing
    a numerical outcome between two independent groups. Supports two-sided and
    one-sided alternatives using histogram-based approximate ranks and an
    asymptotic p-value with tie and continuity corrections.
  • Standardized mean difference analysis: Added the standalone
    standardized_mean_difference analysis, extending the underlying support
    noted in 1.1.0. Computes pooled-variance Cohen's d for every unique pair of
    eligible categories, excluding missing values and groups below the privacy
    minimum.
  • Combined descriptive quartiles: Descriptive statistics now report
    histogram-based estimates of q1, q2, q3, and median across datasets.
    Per-dataset quartiles remain exact; combined quartiles are estimates.

K-means & Reusable Cluster Preprocessing

  • Cluster-count selection: Added automatic elbow selection over a configurable
    range of cluster counts alongside manual selection.
  • Multiple initializations: Added multi_start_random_range, which retains
    the fit with the lowest global inertia, with configurable initialization count
    and random seed.
  • Expanded reporting: Results now include cluster centers, textual profiles,
    size intervals, initialization details, elbow diagnostics, and warnings for
    non-convergence and empty clusters. Observation counts are reported as
    intervals; individual cluster assignments are not returned.
  • Reusable cluster covariates: Added kmeans_cluster_creator preprocessing.
    The reusable_preprocessing payload from a K-means analysis reuses stored
    centers to create a categorical variable for downstream analyses without
    refitting. Replay requires matching source data context and cluster variables.

Mixed-effect Model Improvements

  • Composite grouping: Linear, binary generalized, and ordinal generalized
    mixed models now accept one or two grouping variables. Two variables form one
    composite grouping factor with a single random-intercept variance component.
  • Richer GLMM results: Binary and ordinal GLMM results now expose standard
    errors, z-statistics, p-values and display values, confidence intervals,
    Laplace log-likelihood, AIC, and BIC.
  • Grouping validation: Added checks for duplicate or missing grouping
    variables and for the presence of at least one fixed-effect covariate.

Analysis API & Preprocessing

  • Unified analysis requests: Added POST /analysis with separate source
    inputdata, ordered preprocessing, and algorithm sections. Source columns
    are declared in inputdata.variables; final algorithm inputs can include
    columns created by preprocessing.
  • Separate discovery endpoints: Input-data, preprocessing, and algorithm
    specifications are exposed through GET /specifications/inputdata,
    GET /specifications/preprocessing, and GET /specifications/algorithms.
  • Ordered preprocessing: Steps are validated and executed in request order,
    with variable and metadata changes propagated to subsequent steps and the
    selected algorithm.
  • Rule-based categorical columns: Added categorical_column_creator to
    derive categories from filter rules, with an optional default category.
    Rows matching multiple rules are rejected.
  • Aggregation-aware preprocessing: Preprocessing steps can declare and use
    aggregation-server support, and execution strategy selection accounts for
    preprocessing requirements.
  • Filter validation: Strengthened validation of filter structure, operators,
    value types, and referenced source variables.

Runtime, Validation & Tooling

  • uv migration: Replaced Poetry with uv in the root project and
    exadata-validator, including lockfiles, Docker builds, CI workflows, Invoke
    tasks, and setup documentation. Python remains >=3.10,<3.11.
  • Reduced copying: Removed redundant data copies in numerical computations
    and worker execution paths.
  • Expanded coverage: Added or updated standalone and production-environment
    tests for the new statistics, mixed models, K-means reporting and replay,
    preprocessing, filters, and analysis request validation.
  • Documentation and development tools: Updated the API guide, algorithm
    documentation, and scaffold/validation tools for the analysis request contract
    and uv workflow.

Configuration

📅 Schedule: (UTC)

  • Branch creation
    • At any time (no schedule defined)
  • Automerge
    • At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.


  • If you want to rebase/retry this PR, check this box

This PR was generated by Mend Renovate. View the repository job log.

@renovate renovate Bot added the dependencies label Oct 1, 2026
@renovate renovate Bot changed the title chore(deps): update dependency madgik/exareme2 to v1.2.0 (federation override) chore(deps): update dependency madgik/exareme2 to v1.2.1 (federation override) Oct 4, 2026
@renovate
renovate Bot force-pushed the renovate/federations-madgik-exareme2-1.x branch from 910ecb6 to 8880bb7 Compare October 4, 2026 22:13
@renovate
renovate Bot force-pushed the renovate/federations-madgik-exareme2-1.x branch from 8880bb7 to 8e22040 Compare October 26, 2026 21:29
@renovate
renovate Bot force-pushed the renovate/federations-madgik-exareme2-1.x branch from 8e22040 to 2fc4d08 Compare November 3, 2026 09:48
@jdaln

jdaln commented Nov 3, 2026

Copy link
Copy Markdown
Contributor

@KFilippopolitis marking this as draft since I first need to retest fed-z, my bad

@jdaln
jdaln marked this pull request as draft November 3, 2026 10:16
@renovate
renovate Bot force-pushed the renovate/federations-madgik-exareme2-1.x branch from 2fc4d08 to 4a31664 Compare November 17, 2026 01:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants