Nuée 0.9.9

Documents that cite the project's results and your references, written as Word, PDF or HTML, or compiled by Quarto; the gain over the empty model; and the right coefficients with missing values.

Install

Windows

1. Download Nuee-Setup-0.9.9.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version; from 0.9.1, a click in the status bar installs it.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

Linux

• Debian, Ubuntu and derivatives: download Nuee-0.9.9-amd64.deb and run sudo apt install ./Nuee-0.9.9-amd64.deb. apt brings R and the R packages Nuée needs.
• Other distributions: use Nuee-0.9.9-x86_64.AppImage (needs libfuse2) or Nuee-0.9.9-x64.tar.gz (unpack, then ./nuee).
  • Install R and its packages first: install.packages(c("httpuv", "jsonlite", "openxlsx", "haven", "readxl", "foreign", "nlme", "mgcv", "MASS", "rpart", "cluster")).
  • See the README.
• Older R: Nuée uses the computer's R. With an R older than 4.4 (Ubuntu 22.04 and 24.04, Debian 12), .rds and .RData files don't open, because reading one could run code (CVE-2024-27322).

New in 0.9.9

• Documents: File › New document writes an article, a report or notes in Markdown, beside a preview that follows as you type. Start from a blank page, an article (introduction, methods, results, discussion) or an analysis report. The document is saved with the project, and Undo leaves it alone.
  • The project's results, cited: {{M1.Q2}}, {{M1.R2Y:%}}, {{M1.vip.sugar}}, {{M5.coef.year}} (estimate, interval and p), a plot as it is in its window with its caption ({{M1.scores | Scores of the sites}}), a table of the model report ({{M1.table.resume}}) or the Methods text ({{M1.methodes}}).
    • Type {{ to see the models, then what each offers with its value now.
    • Numbers follow the document's language (0.65 or 0,65). Plots and tables are numbered, and {{ref:label}} gives their number.
    • A model not computed yet is computed when the document cites it.
  • Citations, as in Quarto: [@wold2001], [see @wold2001, p. 33; @eriksson2006], @wold2001 in the sentence, [-@wold2001] without the author. The list of references comes at the end, or where the document puts {{bibliographie}}, author–year or numbered.
    • References come from BibTeX or CSL-JSON files (exported from Zotero, Mendeley, JabRef…), or are pasted.
    • Or from Zotero open on the computer (Zotero 7 or later, with Settings › Advanced › Allow other applications on this computer to communicate with Zotero ticked): as you type @, or in Cite › Insert a citation, in the whole library or one collection. With Better BibTeX, Cite › Pick in Zotero opens Zotero's own picker.
    • The document keeps a copy of what it cites: sent without the .bib, it still cites it.
  • Compile: Word, PDF or HTML, or a Quarto file (.qmd) with its plots and a .bib, which Quarto compiles when it is installed: Word, PDF (Typst) or HTML, in any journal's CSL style.
  • Each document opens in a tab of its own, or in a window. The spell checker works in French and English.
• Gain over the empty model: when the centres are recomputed without the held-out group (the default validation), a PLS or OPLS gives the Q² of the empty model, where each held-out group is predicted by the mean of Y over the other observations.
  • It also gives the model's gain over it, (Q² − Q² empty) / (1 − Q² empty): the share of the empty model's prediction error that the model avoids.
  • With groups that differ (sites, departments), the empty model's Q² is below 0, and a negative Q² can hide a model that does better than it: the gain says so.
  • It is in the Summary, the Methods text, the model report and Export to R; a hierarchical model's honest Q² has its own.
• Fixed: with missing values in X, the coefficients of a PLS or OPLS (scaled and in the data's units), the w* weights and their confidence intervals were off by a few percent as a rule.
  • W* is now computed component after component, as in .usp files. On a real project of 35 models with missing values, the coefficients and their intervals now match those of the software that wrote the file (difference < 10⁻⁶), where they differed by up to 0.044.
  • The "file" values of these coefficients in the explorer had the same error: converting the project again fixes them.
  • R², Q², VIP, scores and predictions don't change, and nothing changes without missing values.

The full list is in Aide › Quoi de neuf and CHANGELOG.md.

Checksum

SHA-256 of Nuee-Setup-0.9.9.exe:
feb6984f97bcdce5070afbe810ce668ed819480a48c5f16320d57b84db03bf92

Nuée 0.9.8

A choice of cross-validation, with a sturdier default for every model; the validation of .usp files, which gives back their Q²; and an honest Q² for hierarchical models.

Install

Windows

1. Download Nuee-Setup-0.9.8.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version; from 0.9.1, a click in the status bar installs it.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

Linux

• Debian, Ubuntu and derivatives: download Nuee-0.9.8-amd64.deb and run sudo apt install ./Nuee-0.9.8-amd64.deb. apt brings R and the R packages Nuée needs.
• Other distributions: use Nuee-0.9.8-x86_64.AppImage (needs libfuse2) or Nuee-0.9.8-x64.tar.gz (unpack, then ./nuee).
  • Install R and its packages first: install.packages(c("httpuv", "jsonlite", "openxlsx", "haven", "readxl", "foreign", "nlme", "mgcv", "MASS", "rpart", "cluster")).
  • See the README.
• Older R: Nuée uses the computer's R. With an R older than 4.4 (Ubuntu 22.04 and 24.04, Debian 12), .rds and .RData files don't open, because reading one could run code (CVE-2024-27322).

New in 0.9.8

• A choice of cross-validation: New model › Cross-validation, for a PCA, PLS or OPLS.
  • Centres recomputed, full model's weights is the new default, for every model, imported ones included. Each group is predicted by a model fitted without it, its centres recomputed without the group, so the model knows nothing of that group's Y. Its weights stay the full model's: X's standard deviations don't depend on Y, and nothing is rescaled on fewer observations, so no prediction blows up.
    • On a real project, Q² went down to −8 × 10¹¹ before, when a variable varied at one site only and the sites were validated apart.
    • Every model's Q² changes a little from 0.9.7.
  • Centres and weights recomputed without the held-out group is 0.9.7's validation, stricter. It now has a guard: a variable that keeps less than 10 % of its standard deviation without the held-out group keeps the full model's weight in that round. The Summary says which variable, and without which group.
  • Full model's centres and weights (like SIMCA) validates each component on what the previous ones leave. That is the validation of .usp files, and it gives back their Q²: 441 of 443 models of four real projects, and 773 of 809 PLS models of the test corpus (all of those without missing data), with their Q² per variable and their coefficients' confidence intervals. It is more optimistic, so it isn't the default.
  • RMSEcv, CV-ANOVA, the cross-validated classification, the confidence intervals, the permutation test, autofit and Export to R follow the validation chosen.
• Honest Q² of hierarchical models: a model whose X include scores of other models of the project ($M1.t1…) gets an Honest Q2(cum) in its Summary and in the Methods text. In each cross-validation round, its base models are refitted without the held-out group, so their scores don't use that group's Y.
• Methods text and model report: the Methods text says how the model was validated, and the model report gives it at the top.
• R console: verifier_moteur() validates the models like their files and counts Q² in its status.
• Fixed: a model with a single component had no "=" or "≠" marks against its .usp file in the Summary, and no R2X or Q2 in the explorer.

The full list is in Aide › Quoi de neuf and CHANGELOG.md.

Checksum

SHA-256 of Nuee-Setup-0.9.8.exe:
03f6782244cfc39b9b1ef42cacf57787afe60ebba46405ef9c84b104d1e5c623

Nuée 0.9.7

Models that follow the design of a monitoring study: qualitative variables in cross terms, adjusted means and pairs of levels, a curve per category, random slopes, a model's observations chosen by a condition; columns added from another table and tables summarized by group; coefficients side by side, trend tests for autocorrelated and seasonal series, diversity indices and the power to detect a trend.

Install

Windows

1. Download Nuee-Setup-0.9.7.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version; from 0.9.1, a click in the status bar installs it.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

Linux

• Debian, Ubuntu and derivatives: download Nuee-0.9.7-amd64.deb and run sudo apt install ./Nuee-0.9.7-amd64.deb. apt brings R and the R packages Nuée needs.
• Other distributions: use Nuee-0.9.7-x86_64.AppImage (needs libfuse2) or Nuee-0.9.7-x64.tar.gz (unpack, then ./nuee).
  • Install R and its packages first: install.packages(c("httpuv", "jsonlite", "openxlsx", "haven", "readxl", "foreign", "nlme", "mgcv", "MASS", "rpart", "cluster")).
  • See the README.
• Older R: Nuée uses the computer's R. With an R older than 4.4 (Ubuntu 22.04 and 24.04, Debian 12), .rds and .RData files don't open, because reading one could run code (CVE-2024-27322).

New in 0.9.7

• Qualitative variables in cross terms: a regression (linear, mixed, GLM, GAM) takes cross terms with a qualitative variable, in the Cross terms tab or as a * b in the R formula.
  • year × sector lets the year's slope change from one sector to another; habitat × treatment, a treatment's effect from one habitat to another.
  • The coefficients are those of lm, nlme and mgcv with the same reference category. Without the year's main effect (sector + year:sector), each coefficient is a sector's slope.
  • The Effects view draws one line (or curve) per category, with the table of slopes. The AIC selection, the contributions and the nested models take a cross term in or out whole, with its main effects.
• Adjusted means and pairs of levels: Plots › Adjusted means gives each category's mean adjusted for the model's other terms, with its 95 % interval, and every pair of categories, corrected by Tukey or Holm, as the emmeans package computes them.
  • With a cross term: the pairs within each level of the other variable, or the year's slopes per category and their pairs.
  • With a log or logit link, on the response's scale (ratios, odds ratios). A sentence ready for a report, the Methods text and Export to R.
• A curve per category in a GAM: s(year, by = sector), each curve with its edf, test, effect and thresholds, in mgcv::gam and mgcv::gamm with CAR1.
• Random slopes: (1 + year | site), each site its own trend around the fixed slope, with the slope's variance, its correlation with the intercept and Johnson's (2014) R²; with CAR1, and in mgcv's GLMMs. A singular fit, such as an uncentred year's, is flagged.
• A model's observations chosen by a condition: New model › Observations › "Keep only the observations where…" (year >= 2010 & site != "S2"), for every type of model. It is kept with the model, evaluated again at each fit, given in the Summary, the Methods text and Export to R, and read without being run.
• Add columns from a table: Edit › Add columns from a table… joins another file's columns (the sites' habitat and altitude, the weather by year) by a key of one or more columns. Repeated keys are refused, unmatched observations listed; Ctrl+Z removes them.
• Summarize by group: Edit › Summarize by group… makes a new project with one row per group (sector and year), with the mean, sum, median, minimum, maximum, standard deviation or count of the variables chosen.
• Coefficients side by side: Prediction › Compare models puts several regressions in columns, a row per term, then n, AIC, R², the random effects, φ and the curves' edf; Ctrl+C copies it as a table for Word or Excel.
• Trend tests: in Plots › Series › Trends, Hamed and Rao's (1998) test for autocorrelated series, and the seasonal Kendall test (Hirsch, Slack and Smith, 1982) for samples taken several times a year.
• Diversity indices: richness(), shannon(), simpson() and pielou() in New variable formulas, over species columns (shannon(Ablette:Vandoise)), as vegan computes them.
• Trend power: Plots › Trend power says how many years it takes to detect a trend (per year, or in % per year) with this many sites counted every year, from a linear or mixed model's variances (CAR1 included), with a simulation that checks it.
• Changes and fixes:
  • A category a regression didn't see in its fit (absent from its observations, or from the training groups in cross-validation, as sectors validated by site) was predicted as the reference category, without a word. It gets no prediction now, and those rows are counted and explained; Q2 can change from 0.9.6.
  • In Block contributions (AIC), a main effect now leaves with the cross terms that contain it, so the ΔAIC of a model with cross terms change from 0.9.6.
  • In Export to R, a name from the project (a variable, a model, a file…) that held a line break could break out of a comment of the script and run when the script was run. Such names are now written on a single line; this matters for a project received from someone.
  • A change the engine refused (a New variable formula that doesn't compute) no longer leaves an empty step for Ctrl+Z.

The full list is in Aide › Quoi de neuf and CHANGELOG.md.

Checksum

SHA-256 of Nuee-Setup-0.9.7.exe:
214db1f1131549f5e4aeefcdeac5eb93a215f24f466e7e9da13174516afe6fa2

Nuée 0.9.6

Monitoring analyses, one step further: the BACI mixed model in one click, trend tests in Series, Friedman's test, "Fit with CAR1", quantile residuals for counts, and sampling dates wherever a time is asked for. The one-click update also no longer shows the previous version's interface, nor a black window.

Install

Windows

1. Download Nuee-Setup-0.9.6.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version; from 0.9.1, a click in the status bar installs it.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

Linux

• Debian, Ubuntu and derivatives: download Nuee-0.9.6-amd64.deb and run sudo apt install ./Nuee-0.9.6-amd64.deb. apt brings R and the R packages Nuée needs.
• Other distributions: use Nuee-0.9.6-x86_64.AppImage (needs libfuse2) or Nuee-0.9.6-x64.tar.gz (unpack, then ./nuee).
  • Install R and its packages first: install.packages(c("httpuv", "jsonlite", "openxlsx", "haven", "readxl", "foreign", "nlme", "mgcv", "MASS", "rpart", "cluster")).
  • See the README.
• Older R: Nuée uses the computer's R. With an R older than 4.4 (Ubuntu 22.04 and 24.04, Debian 12), .rds and .RData files don't open, because reading one could run code (CVE-2024-27322).

New in 0.9.6

• BACI mixed model in one click: in Plots › Statistical tests, Before/after (BACI), "Fit the mixed model" fits y ~ period × impact + (1 | unit), with a CAR1 autocorrelation in time if wanted.
  • It adds the 0/1 period and impact variables to the project.
  • The BACI effect (the period × impact coefficient, its 95 % interval and p) comes as a sentence under the t-test's. It accounts for the units and the autocorrelation.
  • The model joins the project with its Methods text (Stewart-Oaten et al., 1986), and Export to R redoes it.
• Trends in Series: "Trends" gives, for each line, the Mann–Kendall test (Kendall's τ, S and its p) and Sen's slope per year with its 95 % interval. The p-values are corrected over the lines (Benjamini–Hochberg or Holm).
• Friedman's test: when a measurement is paired by a key between three or more levels (sectors, by the year), Friedman's test with Kendall's W now comes before the table of pairs. With "within each" site, there is one test per site, Holm-corrected.
• Fit with CAR1: in Plots › Residual autocorrelation, a button fits the same model with a CAR1 autocorrelation within its random effects' groups (or a finer level within them), to compare by AIC.
• Quantile residuals: for Poisson, negative binomial, binomial and gamma models, the Residuals view offers randomized quantile residuals (Dunn & Smyth, 1996). They are normal when the model is right, where residuals of counts are not even then. They are also in the model report and Export to R.
• Dates as time: sampling dates (2019-05-14, 14/05/2019, day first) are read as decimal years wherever a time is asked for: Series, lags, BACI, residual autocorrelation, and CAR1 with its Export to R. Series and BACI show them as dates. A column of repeated dates can now be the time.
• Fixes:
  • After a one-click update, Nuée could show the previous version's interface for a while: it showed the old number and said it was the latest. The first start of a new version now empties that cache, and the interface is read again at each load.
  • The one-click update no longer opens a black window while it waits. This applies from the next update on, because the update to 0.9.6 still runs the previous version's code.

The full list is in Aide › Quoi de neuf and CHANGELOG.md.

Checksum

SHA-256 of Nuee-Setup-0.9.6.exe:
37ebdcf9375b06a7f4ab20f9059db35b0626de6d9d943d94338e3de29b18cb89

Nuée 0.9.5

Model selection and long-term monitoring: an AIC selection that suggests a model and fits it in one click, paired tests on long data, before/after comparisons with controls (BACI), series over time, the previous year's values in a formula, the residuals' autocorrelation and likelihood-ratio tests of nested models. 0.9.4 was not published on its own: its paired tests are part of this version.

Install

Windows

1. Download Nuee-Setup-0.9.5.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version; from 0.9.1, a click in the status bar installs it.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

Linux

• Debian, Ubuntu and derivatives: download Nuee-0.9.5-amd64.deb and run sudo apt install ./Nuee-0.9.5-amd64.deb. apt brings R and the R packages Nuée needs.
• Other distributions: use Nuee-0.9.5-x86_64.AppImage (needs libfuse2) or Nuee-0.9.5-x64.tar.gz (unpack, then ./nuee).
  • Install R and its packages first: install.packages(c("httpuv", "jsonlite", "openxlsx", "haven", "readxl", "foreign", "nlme", "mgcv", "MASS", "rpart", "cluster")).
  • See the README.
• Older R: Nuée uses the computer's R. With an R older than 4.4 (Ubuntu 22.04 and 24.04, Debian 12), .rds and .RData files don't open, because reading one could run code (CVE-2024-27322).

New in 0.9.5

• AIC selection: Plots › AIC selection (also under each regression in the explorer) compares by AICc every combination of a model's groups of terms (its blocks, otherwise its variables), from the null model to the full one.
  • Each combination is fitted by ML on the same observations, with the model's random effects, autocorrelation, effort and curves. A cross term enters only with its main effects. Beyond 8 groups, a backward elimination.
  • The view suggests the best model and, when it differs, the most parsimonious within ΔAICc ≤ 2. It shows each group's importance and the best models with their Akaike weights.
  • Fit this model, on the suggestion or on any row, makes it a new model, fitted on the full model's rows. The Methods text describes the selection, the model report can include it, and Export to R redoes it.
• Paired tests on long data: in Plots › Statistical tests, Two variables mode, a measurement and a qualitative variable can be paired by a key (Pair by: the year).
  • Two sectors are compared on the years both have, by the paired t-test or Wilcoxon's signed-rank test. The data stay one row per sector and year.
  • The table can list every pair of sectors, with Holm's correction, or only the pairs within each site.
  • Two measurements of the same rows (before and after) are paired by the observation.
  • In Between classes mode, two classes paired by a key are compared variable by variable, with the volcano plot and the corrections.
• Before/after with controls (BACI): a fourth mode of Plots › Statistical tests. At each time, the gap between the impact units and the controls. The gaps after the event are compared with those before, by Welch's t, Student's t or Mann–Whitney, as a sentence ready for a report, with a plot of the gaps.
• Series: Plots › Series shows a variable over time, one line per sector and, optionally, one small plot per site, with shared axes. A missing year breaks a line; optionally, the mean of each site's lines, a log scale and events.
• Values from another year: previous(x, year, sector), following(x, year, sector) and growth_rate(x, year, sector) in a New variable formula (precedent, suivant and croissance in French): year n against year n − 1, or density dependence.
• Residual autocorrelation: Plots › Residual autocorrelation shows a regression's residual autocorrelation in time within each group, as nlme::ACF computes it. A model with CAR1 also shows its normalized residuals, which show none left when CAR1 is enough; for another model, the note says when to try CAR1.
• Nested models: under the AIC table of Prediction › Compare models, the likelihood-ratio test between nested regressions, as nlme's anova(m1, m2).
• Fixes:
  • The normalized residuals of a model with CAR1 (Residuals view, exports, Export to R) were wrong when the rows weren't sorted by group, then by time. The fit itself (coefficients, φ, AIC) was not affected.
  • The views computed from a project's data (statistics, statistical tests) are recomputed when its data change.
  • Mann–Whitney's test now computes the same way on every R version. With ties, its p-value therefore changes slightly from 0.9.3 on Windows.

The full list is in Aide › Quoi de neuf and CHANGELOG.md.

Checksum

SHA-256 of Nuee-Setup-0.9.5.exe:
f69f67a8c7c6eb78e8326d32555186609db3523ee2dba6b20713ef87a8794e7a

Nuée 0.9.3

Community and classification analyses: redundancy and correspondence analysis, decision trees, linear discriminant analysis and k-means. Additive mixed models with an autocorrelation in time, a curve's thresholds, group variables and imputation of missing values.

Install

Windows

1. Download Nuee-Setup-0.9.3.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version; from 0.9.1, a click in the status bar installs it.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

Linux

• Debian, Ubuntu and derivatives: download Nuee-0.9.3-amd64.deb and run sudo apt install ./Nuee-0.9.3-amd64.deb. apt brings R and the R packages Nuée needs, now including MASS, rpart and cluster.
• Other distributions: use Nuee-0.9.3-x86_64.AppImage (needs libfuse2) or Nuee-0.9.3-x64.tar.gz (unpack, then ./nuee).
  • Install R and its packages first: install.packages(c("httpuv", "jsonlite", "openxlsx", "haven", "readxl", "foreign", "nlme", "mgcv", "MASS", "rpart", "cluster")).
  • See the README.
• Older R: Nuée uses the computer's R. With an R older than 4.4 (Ubuntu 22.04 and 24.04, Debian 12), .rds and .RData files don't open, because reading one could run code (CVE-2024-27322).

New in 0.9.3

• Redundancy analysis (RDA): Nouveau modèle › RDA asks how much of a table of responses (species abundances, measurements) explanatory variables explain, and how.
  • The responses can be transformed by Hellinger, chord or log(x + 1), and covariates can be removed first (partial RDA).
  • Results: triplots in scaling 1 or 2 and the adjusted R².
  • Permutation tests of the model, of each axis and of each term, within sites if needed, with a fixed seed.
  • Forward selection and variation partitioning between blocks of variables.
  • The results are those of the vegan package.
• Correspondence analysis (CA) of a table of counts or abundances, with contributions and cos².
• Decision trees (rpart): a regression or classification tree on the raw variables, so its thresholds read in their units.
  • Missing values are handled by surrogate variables.
  • Pruned by the 1-SE rule through a cross-validation by site, and validated by predicting each site with a tree rebuilt without it.
  • Results: the tree drawn, the rules of its leaves, the importance of the variables, and the stability of the first split over 200 bootstrap draws of the sites.
  • With little data, the minimum per leaf is lowered so that the tree can split.
• Linear discriminant analysis (MASS::lda): discriminant scores and coefficients, a classification cross-validated by groups, and posterior probabilities.
• k-means: clusters on the scaled variables, with silhouettes and the choice of k. Groups → variable makes the clusters a qualitative variable of the project.
• Like the other models, these analyses have a Methods text, export tables, a PDF or Word report and an Export to R, checked against Nuée's results before it is written. Trees and LDA predict test sets and new observations.
• Additive mixed models: a GAM or GLM with random effects and an autocorrelation in time (CAR1) is fitted by mgcv::gamm.
  • The autocorrelation can run within a finer level than the last random effect: the sector within its site.
  • Each curve can be given its basis dimension (k).
• A curve's thresholds: the Effects view shows a curve's effect and slope with 95 % simultaneous bands (2000 draws of the coefficients), and the ranges where the slope clearly rises or falls. They are also in the report and in Export to R.
• Concurvity of each curve in an additive model's Summary.
• Group variables: group_mean(x, sector) and group_deviation(x, sector) in a New variable formula (moyenne_groupe, ecart_groupe in French).
• Imputation of missing values: regularized iterative PCA, as missMDA::imputePCA does. The number of components is chosen by generalized cross-validation of the standardized variables.
• Fixes:
  • Edit model keeps the blocks and modifiers of cross terms.
  • The one-click update now updates an all-users installation outside Program Files in place, instead of installing a second copy.

The full list is in Aide › Quoi de neuf and CHANGELOG.md.

Checksum

SHA-256 of Nuee-Setup-0.9.3.exe:
e2e32ebf2561f67480b36b3ed095647c5c80ac107188b511f149924959784169

Nuée 0.9.2

Regression models compared by AIC: linear, mixed, generalized and additive, with an R formula box, predictions, exports and reports; and statistical tests between two variables, with a correlation matrix.

Install

Windows

1. Download Nuee-Setup-0.9.2.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version; from 0.9.1, a click in the status bar installs it.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

Linux

• Debian, Ubuntu and derivatives: download Nuee-0.9.2-amd64.deb and run sudo apt install ./Nuee-0.9.2-amd64.deb. apt brings R and the R packages Nuée needs, now including nlme and mgcv for the regressions.
• Other distributions: Nuee-0.9.2-x86_64.AppImage (needs libfuse2) or Nuee-0.9.2-x64.tar.gz (unpack, then ./nuee), with R and its packages installed first: install.packages(c("httpuv", "jsonlite", "openxlsx", "haven", "readxl", "foreign", "nlme", "mgcv")). See the README.
• Nuée uses the computer's R. With an R older than 4.4 (Ubuntu 22.04 and 24.04, Debian 12), .rds and .RData files don't open, because reading one could run code (CVE-2024-27322).

New in 0.9.2

• Linear and mixed models: Nouveau modèle has Régression linéaire and Modèle linéaire mixte (nlme), built from the Variables tab like the other models. Random effects are nested (site, then sector within the site), from an observation ID or a qualitative variable, with an optional autocorrelation in time (CAR1). Coefficients and variances by REML, AIC by ML, marginal and conditional R². Cross-validation is by site by default: each site is predicted as a site never seen.
• Generalized and additive models (mgcv): Poisson or negative binomial for counts, binomial for presence-absence, gamma for positive values, Tweedie for densities with zeros. An effort can enter as an offset, random effects give GLMMs and GAMMs, and the variables ticked in the new Courbe column enter as curves, for an effect that rises then falls or a threshold. A Poisson model warns when its counts are overdispersed.
• Results: the coefficients with their 95 % intervals (multiplicative effects or odds ratios for log and logit links), a new Effets view with each variable's effect curve, the residuals, and the AIC change when each block of variables is removed.
• Comparing models by AIC: Prédiction › Comparer les modèles gives AIC, ΔAIC and Akaike weights between the regressions that can be compared (the same response on the same observations, with likelihoods that compare), next to every model's cross-validated Q², PLS included.
• R formula: Nouveau modèle › Général shows the model as R writes it, y ~ a * b + log10(c) + s(d) + offset(log(effort)) + (1 | site/secteur); edit it and apply it to set the tabs. R reads it without ever running it, and it isn't kept, so a project someone sends you can't carry code.
• Like the other models: regressions predict observations and files (extrapolations flagged), export their tables, make a PDF or Word report, and export to R: a script or R Markdown / Quarto report that refits them with lm, nlme or mgcv, checked against Nuée's results before it is written.
• Statistical tests: Graphiques › Tests statistiques (formerly Tests par variable) adds two modes. Deux variables picks the test from the two variables' types and checks its conditions (chi-square or Fisher's exact test; Welch's t, ANOVA or Kruskal–Wallis with pairwise comparisons; Pearson or Spearman), then gives the result as a sentence ready for a report, with its effect size. Matrice des corrélations shows every pair of quantitative variables, with p-values corrected for the number of pairs.

The full list is in Aide › Quoi de neuf and CHANGELOG.md.

Checksum

SHA-256 of Nuee-Setup-0.9.2.exe:
632145cde7ecefb539782348a5950716fc95a61e89919d513a115082185cee6e

Nuée 0.9.1

SIMCA projects refitted as they were built (cross-validation groups, block scaling, modifiers), a cross-validation that no longer inflates Q², a test set, one-click updates, and Linux packages.

Install

Windows

1. Download Nuee-Setup-0.9.1.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps. From this version on, a click in the status bar installs the next version.

Linux (new)

• Debian, Ubuntu and derivatives: download Nuee-0.9.1-amd64.deb and run sudo apt install ./Nuee-0.9.1-amd64.deb. apt brings R and the R packages Nuée needs.
• Other distributions: Nuee-0.9.1-x86_64.AppImage (needs libfuse2) or Nuee-0.9.1-x64.tar.gz (unpack, then ./nuee), with R and its packages installed first. See the README.
• Nuée uses the computer's R. With an R older than 4.4 (Ubuntu 22.04 and 24.04, Debian 12), .rds and .RData files don't open, because reading one could run code (CVE-2024-27322).

New in 0.9.1

• SIMCA projects as they were built: models from .usp files keep their cross-validation groups. These can be one group per value of an observation ID (site, sector, year…) or the file's own groups, instead of 7 groups in turn. Block scaling (1/√K) and modifiers are applied as stored. Nouveau modèle has the same options: cross-validation groups by an ID or a variable, and Bloc and Modificateur columns. On a 132-model project compared with SIMCA 17, R²X and R²Y now match for every model.
• More transformations: Transformation › personnalisée… offers log10, ln or −log10 of (C1·x + C2), exp, linear, a logit between two bounds, or the power (C1·x + C2)^C3, with the formula shown. Cross terms can have three factors (A × B × C, A × A × A).
• Cross-validation without leaks: each group is now predicted by a model fitted without it, centring and scaling included. Until now, Q² was inflated when there are few observations: an OPLS-DA of 6 random observations had a Q² of 0.98. Every model's Q² changes: a little in general, a lot with few observations. RMSEcv, CV-ANOVA and the number of components chosen automatically change with it. The Q² of a model imported from a .usp file no longer equals the one stored there.
• Test set: observations held out of the model and of the cross-validation, chosen by hand, at random or by Kennard–Stone, then predicted (RMSEP, bias, SEP, external Q², or the classification of a PLS-DA).
• Projects as .nuee: double-click one to open it, or drop files on the window. Saving never leaves a file half-written, and Nuée asks before overwriting a file changed elsewhere. Recovery copies are kept until recovered or set aside.
• Security: the R engine answers only Nuée's own windows. A file received from someone (project, R file, export to R) can no longer run code.
• Field data: a text cell (<0.5, <LOD, nd, #N/A…) in a column of measurements is read as missing, and Nuée says so. Excel dates stay dates.
• Export to R: the script holds only the exported model's calculation, in plain R with no package (about 170 lines for a PLS), or the whole session in one script.
• And:
  • A variable clicked in one plot is highlighted in all of the model's plots.
  • Graphiques › iPLS fits each interval of the spectrum on its own.
  • Ctrl+C copies the view's table; Ctrl+K finds a command.
  • The R console shows each fitted model's code.
  • The engine's messages and the full user guide are available in English.
  • IT departments get silent installation and a policy that turns the update check off (docs/deploiement.md).

The full list is in Aide › Quoi de neuf and CHANGELOG.md.

Checksum

SHA-256 of Nuee-Setup-0.9.1.exe:
6663b8fa0299106518af1a49aa1605f630d9af5812f33a44d6c1f239047b5cac

Nuée 0.9.0

Spectra straight from the instrument's files, two more preprocessing steps, wavelength ranges chosen with the mouse, and a correction to OPLS with several Y.

Install

1. Download Nuee-Setup-0.9.0.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

New in 0.9.0

• Import spectra: Fichier › Importer des spectres reads JCAMP-DX (.jdx, .dx, .jcm, every compression form), SPC (Galactic, Thermo; new and old format) and text files with one column per spectrum, one spectrum or several per file, into one project: one row per spectrum, one column per wavelength. A table of sample information (contents, classes, batches) can be joined to them by file name. Spectra on different grids are interpolated onto the first one's, and the status bar says so. A single JCAMP-DX or SPC file also opens with Fichier › Ouvrir.
• Detrend and baseline: two more preprocessing steps (a degree-2 polynomial removed, after Barnes et al. 1989; the lower convex hull removed), and up to three steps per model.
• Wavelength ranges: in Graphiques › Profils (spectres), tick Choisir des zones and drag across the plot to mark ranges, such as the water bands; then Nouveau modèle sans ces zones or avec ces zones seules. Savitzky–Golay now treats each range of neighbouring variables on its own, so its window never reaches across a gap, and the lines break where variables are left out.
• OPLS with several Y, or OPLS-DA with more than two classes: the orthogonal components are now unrelated to Y, as in Trygg and Wold's algorithm. In 0.8.0 they were made orthogonal to the PLS weights instead, which left them slightly correlated with Y. Such models change a little when refitted. Models with one predictive component give exactly the same results as before.
• Reports: the R Markdown and Quarto reports from Exporter vers R were rendered to HTML with rmarkdown and Quarto, and now declare their language.

The readers were checked against the official JCAMP-DX test files and the usual SPC sample files, and the preprocessing against the prospectr R package. The 944 reference models give the same results as in 0.8.0.

Checksum

SHA-256 of Nuee-Setup-0.9.0.exe:
07ef8e4a06e06e08851df402067dc2bc8805f16ce4d2dfd88ce4871715a828fe

Nuée 0.8.0

Your model as a standalone R script or report, the Methods paragraph written for you, tests by variable, class models, and OPLS with several classes.

Install

1. Download Nuee-Setup-0.8.0.exe below.
2. Run it and choose to install for yourself (no administrator rights needed) or for everyone on the computer. It replaces an earlier version.
3. The installer isn't code-signed, so Windows may say it protected your PC: click More info, then Run anyway.

R is included; nothing else to install. Windows 10 or 11, 64-bit. Uninstall from Settings › Apps.

New in 0.8.0

• Export to R, without Nuée: Fichier › Exporter vers R writes a script, or an R Markdown or Quarto report, that refits the model in plain R. It holds the code of the engine functions the model needs (copied with their comments), the data (or reads them from a CSV next to it) and the model, and checks the R version and packages first. The report adds the Methods text, tables, plots and references, and compiles in RStudio (Knit) or with Quarto. Optionally also a renv.lock, the data and fitted model as .RData, and an RStudio project. Nuée refits the script's model before writing it and checks that it gives the same results.
• Methods paragraph: Modèle › Texte pour la section Méthodes writes the paragraph for an article from the model (data, preprocessing, method, cross-validation, R², Q², CV-ANOVA, software), in French or English, with its references.
• Tests by variable: Graphiques › Tests par variable compares every variable between two classes (Welch's or Student's t, Mann–Whitney) or across all of them (ANOVA, Kruskal–Wallis), with the Benjamini–Hochberg correction and a volcano plot.
• Class models: a new model type fits one PCA per class; the Coomans plot shows each observation's distance to two of them, to tell which class it belongs to, several, or none.
• OPLS with several Y, and OPLS-DA with more than two classes, through several predictive components. Models with one predictive component give exactly the same results as before.
• Exported files checked in the real programs (jamovi, PSPP for SPSS, Weka, LibreOffice, pandas for Stata and SAS). This fixed three problems: jamovi now opens Nuée's .omv files; SAS files use transport version 5, which SAS reads with libname xport; dBase files use Windows-1252 with missing values left blank.

Models made with earlier versions give the same results: the 944 reference models were refitted and compared again.

Checksum

SHA-256 of Nuee-Setup-0.8.0.exe:
9b3994bdd7872fca3abe9ca31448f29723f924fd4295f4018eddcce75a9bf5c4