Skip to contents

Changelog

Source: NEWS.md

rENA 0.4.18 (development version)

  • rena-wasm: detectNesting(rows, codes, { threshold = 0.95, minRows = 5, codebook = null }) suggests nested codes (a code B that is a subcode of A, B ⊆ A) from a coded table, so the webtool can propose a codebook hierarchy instead of having it rebuilt by hand. Three signals, strongest first: a codebook ([{ name, parent }], confidence high), the sub’s name (A > B, A / B, A: B, A - B, A/B, A:B, A.B, A_B, A (B) when A is a code; medium) and the data (rows with both / rows with the sub at least threshold over at least minRows lines, with the parent occurring somewhere without the sub; medium with a matching name, low alone). One pair per sub (strongest signal, then highest containment, then the parent with most rows), other parents that pass the data threshold as alternatives, lines / withParent and notes on every pair, chains flattened to the top-most ancestor, codes on exactly the same rows reported as duplicates, never as a pair. Async: the data signal (counts, co-occurrence, pairs by containment, duplicates) is libqe’s nesting_candidates() (@qe-libs/libqe-wasm >= 0.2.6, loaded on first call, or passed as opts.qe), since nesting is applied to the code matrix before accumulation and so belongs to every model; the codebook and name signals and their combination are JS here. Tests in wasm/test/detect-nesting.test.js (LIBQE_WASM_DIR=<libqe>/wasm tests against an unreleased libqe); documented in wasm/README.md.

rENA 0.4.17

Released with qe-ena 0.1.11 (tag python/v0.1.11) and rena-wasm 0.1.18 (wasm/v0.1.18). libena’s C++ is unchanged since 0.4.16, so there is no libena release this cycle. ENA.jl’s matching fix ships with ENA.jl 0.1.5 (tag julia/v0.1.5), released on its own once its ENA_jll build is available.

  • model(), center() and optimize() follow ena.make.set(center.align.to.origin = TRUE), the webtool’s rule, for units whose network is empty: with exclude_zero_networks = TRUE, now model()‘s default for every kind of set, an all-zero unit stays at the origin. center() centers the other units on their own mean and leaves the empty ones at zero (before, it subtracted that mean from every unit, so an empty network landed at minus the mean, as far from the centroid as any real unit). optimize() gains exclude_zero_networks (default TRUE): empty units are left out of the node fit and get zero centroids, and for unordered sets the nodes and centroids are shifted so the fitted units’ mean centroid is at the origin, as the legacy solver does; ordered sets are fit on the other units without that shift, as ONA 0.1.1 did. On RS.data with units including GroupName (two empty networks of 96), the piped model now matches ena.make.set() to 1e-10 in points, nodes, centroids and eigenvalues; before, points differed by up to 0.5 and nodes by up to 1.0. Ordered models change the same way: an empty unit’s point moves from minus the mean to the origin and the rotation from dimension 3 on, matching ONA 0.1.1 (ona’s parity fixture). Sets without empty units are unchanged. exclude_zero_networks = FALSE reproduces center.align.to.origin = FALSE.
  • rena-wasm, qe-ena and ENA.jl follow the same zero-network rule as model() above, so all four agree with ena.make.set(center.align.to.origin = TRUE) and ONA 0.1.1 on data with units whose network is empty: such a unit stays at the origin, is left out of the centering mean and the node fit, and has a zero centroid; unordered models shift the nodes and centroids so the fitted units’ mean centroid is at the origin, ordered models are fit on the other units without that shift. rena-wasm: fit()’s centerAlignToOrigin (default true) now applies to ordered (ONA) models too, which used to shift an empty unit to minus the mean as the old ona::model() did, and the node fit leaves empty units out; centerAlignToOrigin: false is still center.align.to.origin = FALSE for every kind of model. The rs_zero ONA fixtures (wasm/test/fixtures/ona-model-rs.json) were refrozen from this rule with wasm/test/fixtures/ona-fixtures-rs.R, which now runs against the rENA and ona checkouts. qe-ena and ENA.jl already centered on the non-empty units; their node fit now leaves the empty units out as well (zero centroids, the unordered shift). New parity tests against model() on RS.data accumulated by UserName, Condition and GroupName (two empty units unordered, one ordered; python/tests/fixtures/zero-networks.R, the JSON shared with julia/test/fixtures/) match points, nodes, centroids, eigenvalues and variance to 1e-10. Models without empty units are unchanged in all three.
  • rena-wasm requires @qe-libs/libqe-wasm >= 0.2.5 (was >= 0.1.6, and its lockfile still pinned 0.1.6), so it always loads the libqe-wasm that is built without dynamic execution (0.2.4 and later), as 0.1.17’s CSP-safe loading needs.
  • rena-wasm 0.1.17 (released on its own, tag wasm/v0.1.17) is linked with -sDYNAMIC_EXECUTION=0, so libena’s Embind glue no longer builds function invokers with new Function. It loads under Cloudflare Workers and any Content Security Policy without unsafe-eval (the ENA MCP server’s Worker build needs this, together with libqe-wasm 0.2.4). Results are unchanged. R, qe-ena, ENA.jl and the C++ are unchanged.

rENA 0.4.16

Released with qe-ena 0.1.10 (tag python/v0.1.10), rena-wasm 0.1.16 (wasm/v0.1.16) and libena 0.4.16 (conan/v0.4.16). rena-wasm’s README examples run in the browser on qe-libs.org, and libena’s Conan docs gain a main page.

  • ENA.jl 0.1.4 (released on its own, tag julia/v0.1.4) fits ENA models end to end, as rENA and qe-ena do (before, it had only the libena kernels). ENA.accumulate(data, units, conversations, codes; window_size, window_forward, binary) is the stanza-window accumulation (TMA.jl’s accumulate_stanza) on any Tables.jl table, returning an ENAAccumulation. ENA.fit(accumulation; dims, norm, rotation, self_weight) returns an ENAModel (points, nodes, centroids, rotation, line weights, variance, …): sphere norm and centering from LibQE.jl, rotation by SVD (default), MeansRotation or GeneralizedRotation (target, covariates with interactions, y, select_2_groups) through libena’s ena_svd / means_rotation / generalized_means_rotation, and libena’s node positions. It also models TMA.jl’s TMA.accumulate results: ordered networks with ONA’s directed node positions, self-connection columns by their adjacency key. move_nodes_to_unit_circle(model) moves a fitted model’s nodes, and the new kernel binding move_nodes_to_unit_circle_by_key() handles self-connection layouts. On RS.data the accumulation and the SVD, means and four generalized models match rENA’s ena.accumulate.data() / ena.make.set() to 1e-10 (julia/test/fixtures/ena-rs.R), and models of tma accumulations match rENA’s model() (julia/test/fixtures/tma-models.json). ENA.jl now depends on TMA.jl (>= 0.1.6, for TMAAccumulation), LibQE.jl and Tables.jl.
  • ENA.jl 0.1.3 (released on its own, tag julia/v0.1.3) loads its C++ library from ENA_jll, built by cranqe with BinaryBuilder for Linux (x86_64, aarch64) and Windows (x86_64), so Pkg.add("ENA") needs no compiler there; it pins ENA_jll to its own version (=0.1.3). Where ENA_jll has no build (macOS for now), ENA.jl loads a local CMake build from julia/lib/ as before; in a checkout, julia/dev/local_jll.jl develops a stand-in ENA_jll pointing at that build (julia/README.md: Development). ENA.jl 0.1.3 also carries this cycle’s libena changes above (*_by_key, self_weight).
  • qe-ena models TMA accumulations, as rENA’s model() models tma’s in R: ENA().fit() accepts qe-tma’s TMAAccumulation (tma.accumulate()), placing ordered networks with ONA’s directed node positions and unordered networks with self-connection columns by their adjacency key; new self_weight (default 2, = R) and ordered_ / self_weight_ on the model, which move_nodes_to_unit_circle() follows. Matches rENA’s model() (SVD) on tma’s accumulation to ~1e-15 (python/tests/fixtures/tma-models.R). qe-tma no longer models (it imported qe-ena for node positions); qe-ena reads the accumulation’s fields and does not require qe-tma. Standard ENA fits are unchanged.
  • Requires libqe >= 0.2.1 (diagonal adjacency helpers) and tma >= 0.3.9 (accumulate(self_connections)); scripts/sync-headers.sh vendors libqe 0.2.1 and libtma 0.3.9, and the libena Conan recipe requires libqe >= 0.2.1.
  • ena.plot.network() plots networks with self-connections (p(p+1)/2 weights from accumulate(self_connections = TRUE), upper triangle with the diagonal). Each self weight is a filled circle on its node, as wide as an edge of the same weight and coloured the same way (positive/negative for subtractions); it counts once toward the node’s size, and a node is never drawn smaller than its circle. Works on both backends; the qeviz backend needs qeviz with undirected self-connection support (sandbox/qeviz feat/undirected-self-connections). namesToAdjacencyKey() gains diagonal for the self-connection layout. Networks without self-connections plot as before.
  • Unordered models can include self-connections: accumulate(self_connections = TRUE) (passed to tma::accumulate()) adds each code’s self-connection as a connection column (A & A, A & B, B & B, …), and model() takes it through normalization, rotation (SVD, means and generalized) and node positions, where it counts self_weight times (default 2), as in ordered models. Unordered node positions now come from libena’s key-driven solver using the set’s adjacency key – identical to before for the usual choose(n, 2) columns – and rotate() keeps the adjacency key and centering vector it used to drop. Ordered models are unchanged (they always carry the diagonal and keep directed_node_positions()). rena-wasm: accumulate() / fit() take selfConnections; its unordered nodes, custom-rotation centroids, code mask and unit-circle move follow the key. Window tuning and PRIA keep the standard layout. Results match between R and rena-wasm (rs.data: line weights, points and nodes to ~1e-14).
  • Layout-agnostic node optimization in libena: node_weights_by_key(), node_positions_by_key(), node_centroids_by_key() and move_nodes_to_unit_circle_by_key() take a column key (which two codes each connection column connects) instead of assuming a layout. One rule: every connection counts toward each code it connects, a self-connection toward its code self_weight times (default 2), each unit’s weights L1-normalised. With the undirected key it equals the undirected solver; with the ordered key, directed_node_positions() at the same self_weight; it also handles undirected networks with self-connection columns (A & A, A & B, B & B, …), where on symmetric data it equals ONA’s weighting. Key helpers: undirected_connection_key(n, diagonal) / directed_connection_key(n). R: node_positions_by_key() / node_centroids_by_key() (from an adjacency key and the codes); qe-ena, rena-wasm and ENA.jl expose the same functions. Uses libqe’s connection_indices(n, row, diagonal).
  • Ordered (ONA) node optimization counts a self-connection (the diagonal, ground i -> response i) twice in its code’s node weight again: a node’s weight is the sum of its column plus the sum of its row of the unit’s adjacency matrix (in + out), as Zhiqiang Cai defined it in directedENA (2021-10-29). libena had counted it once since May 2026, on the premise that the diagonal was always zero, which it is not. Node positions and centroids of ordered models change (on RS.data, window 4, nodes move ~13% of their spread on average); points, rotation and variance do not. New self_weight (default 2; 1 = the once-counted rule) on model(), optimize(), directed_node_positions() and directed_node_positions_with_ground_response_added(), recorded as set$rotation$self.weight and used by move_nodes_to_unit_circle(); also in libena (directed_node_weights, directed_node_positions, node_centroids, move_nodes_to_unit_circle), rena-wasm (fit({ selfWeight }), rotation.selfWeight), qe-ena and ENA.jl.
  • ena.plot.network(scale.weights = TRUE) no longer fails with invalid hsv color. The thickness, saturation and opacity ranges were still in unscaled units after the network was normalised to max |w| = 1, so colours and opacities went outside [0, 1] (almost any network whose largest weight is under 1 failed, subtraction networks included). The ranges are now scaled with the network, and saturation and opacity are clamped to [0, 1]. Colour and opacity now match scale.weights = FALSE, and edge widths fall within scale.range times the plot multiplier. Before, the edges of a FirstGame - SecondGame network on RS.data would have been up to 44.5 px wide next to 3-10 px nodes. scale.weights = FALSE is unchanged.

rENA 0.4.15

  • Requires tma >= 0.3.8 (scripts/sync-headers.sh vendors libtma 0.3.8), which keeps missed co-occurrences and self-connections through accumulation. Sets from accumulate() now carry self.connection.counts – each code’s ground count times its response count (the ordered network’s diagonal, which unordered networks fold away) – and, with misses, miss.counts and self.miss.counts (the code in the ground, absent from the response), plus their per-line tables in model. They are kept alongside the connection counts and do not change the model. Tests check that they survive model() and leave the fitted model unchanged.
  • accumulate() gains misses (default TRUE), passed to tma::accumulate(); misses = FALSE skips them. Before, rENA always got tma’s default and could not turn them off.
  • rena-wasm: accumulation goes through libtma’s accumulate_unit_networks() (one call per unit and context: fold, per-line weight model, unit sum, self-connection diagonal) and weight_rows() for the per-line rowConnectionCounts, so the JS no longer finalises rows itself. Results are unchanged (identical on rs.data; sqrt within 1e-13). accumulate() and fit() gain selfConnectionCounts (units x codes) and, with misses (default true), missCounts and selfMissCounts – equal to R’s self.connection.counts, miss.counts and self.miss.counts. The new libtma functions (and accumulate_tensor_unit_misses) are now reachable through loadENA().
  • qe-ena: ENA.fit()’s default SVD rotation reports rotation_eigenvalues_ as sdev^2 (= R’s rotation$eigenvalues and rena-wasm), not raw singular values. It now uses libena’s ena_svd instead of numpy.linalg.svd, keeping min(units, connections) axes.

rENA 0.4.14

  • move_nodes_to_unit_circle() and move_nodes_to_unit_circle_with_equal_space() run in libena (inst/include/libena/positions.hpp: unit_circle_nodes(), node_centroids(), move_nodes_to_unit_circle(); the node weighting of node_positions() / directed_node_positions() is factored out into node_weights() / directed_node_weights() for it, with identical results), so every binding shares one implementation. Same signatures and, for unordered sets, the same results. They now also work for ordered sets: the centroids use the directed weighting the fit uses (the R code read only the i < j connections, so its ordered centroids were wrong unless the networks were symmetric). Centroids come from set$line.weights, as the fit’s do; the input set is never modified, the columns keep their classes (the centroids’ used to lose ena.dimension), and an equally spaced move of a set whose nodes are all zero in the plane leaves it unchanged instead of failing. Documented, with examples.
  • The other bindings get the same post-fit call: qe-ena ENA.move_nodes_to_unit_circle(dims=(0, 1), mode="equal") (returns a moved copy) and ena.libena.move_nodes_to_unit_circle(); rena-wasm moveNodesToUnitCircle(model, { dims: [0, 1], mode: 'equal' }) (returns a moved copy) and qe.move_nodes_to_unit_circle; ENA.jl move_nodes_to_unit_circle(line_weights, nodes; dims=(1, 2), mode=:equal, directed=false). Fitting is unchanged.
  • Works with libqe 0.2.0, which drops its copies of the ENA and accumulation code (rENA no longer uses them). qe-ena 0.1.7 allows qe-lib<0.3; rena-wasm allows @qe-libs/libqe-wasm 0.2 (its libqe-wasm parity tests skip the functions 0.2 removed).
  • CI and scripts/sync-headers.sh take tma and libtma from their releases; the fallbacks to tma’s develop branch are gone.

rENA 0.4.13

Bug Fixes & Improvements

  • rENA is a compiled package again. The ENA model code (rotations, generalized means rotation, node positions, ena_correlation, ccd_window) moved out of libqe into rENA as the header-only C++ layer libena (inst/include/libena/, wrapped in src/libena_rcpp.cpp), so libqe can return to generic numerics. Results are unchanged: the code and its tests moved verbatim. Installing from source now needs a C++ toolchain (Rtools on Windows); qe-libs.org serves binaries for Windows and macOS. Requires libqe >= 0.1.9.
  • The other bindings compile libena themselves instead of calling libqe’s copies: qe-ena gets the ena._libena extension (ena.libena; wheels for Linux, macOS and Windows), rena-wasm builds dist/libena.wasm, and the new Julia package ENA.jl (julia/) replaces LibQE.jl’s ENA functions — adding ccd_window and fixing ena_correlation (wrong result shape) and directed_node_positions_combine_pairs (always errored) on the way. libena is also published as a Conan package.
  • Accumulation comes from tma instead of libqe: the R stanza-window wrappers call tma::accumulate_stanza / row_connections / rolling_window_sum (requires tma >= 0.3.6), and qe-ena and rena-wasm compile tma’s libtma accumulation into their own modules (identical results).

rENA 0.4.12

Bug Fixes & Improvements

  • rena-wasm 0.1.11: model.variance is each dimension’s share of the total variance whatever dims a fit keeps, as R’s set$model$variance is regardless of dimensions. It previously divided by the variance of only the returned dimensions, so a default 2-dimension fit always reported shares summing to 100% (58.1% / 41.9% instead of 31.9% / 23.0% on RS.data). Fits that keep every dimension (the webtool’s) are unchanged.

rENA 0.4.11

Bug Fixes & Improvements

  • Requires libqe >= 0.1.6; rena-wasm 0.1.10 requires @qe-libs/libqe-wasm ^0.1.6.

  • GMR (ena.rotate.by.generalized, and rena-wasm’s 'generalized' rotation) is orthonormal on models with a masked (all-zero) connection, through libqe 0.1.6’s complete_rotation fix. One of the SVD axes used to nearly duplicate GMR1, skewing those axes and the variance shares (GMR1 23.7% instead of 29.9% on RS.data with one connection masked). Unmasked models are unchanged.

  • rena-wasm 0.1.10: a model projected into another model’s space (rotationSet) reports each dimension’s share of the total variance, as ena.make.set(rotation.set = ) does with the full rotation matrix. It previously divided by the variance of only the dimensions the rotation set carries (the webtool stores 6), which inflated the percentages (e.g. 41.5% instead of 28.8% on dimension 1).

rENA 0.4.10

New Features

  • skip_sphere_norm(), the piped counterpart of sphere_norm() for fun_skip_sphere_norm: every network is divided by the length of the longest one instead of being scaled to length 1. Pass it as normalize to model() (or ona::model(), which hands it on) to build a model without sphere normalization, including ordered (ONA) models.

Bug Fixes & Improvements

  • rena-wasm 0.1.9 (@qe-libs/rena-wasm) adds the remaining window and rotation options of the R pipeline, each matching R on RS.data:
    • window: Infinity = window.size.back = Inf.
    • Time-based windows: a tensor’s timesCol (parsed like rENA.api’s parse_date), timeUnit and timesEndCol match tma::accumulate() with a time column.
    • Flexible horizons: horizons: { by, rules } builds each unit’s contexts as tma::contexts() does with HOO rules, split on the horizon columns.
    • Custom rotations: rotationSet projects into another model’s space, as ena.make.set(rotation.set = ) does.
    • sphereNorm: false (= norm.by = fun_skip_sphere_norm; for ordered models, ona::model(normalize = skip_sphere_norm)) and, for unordered models, centerAlignToOrigin: false (= center.align.to.origin = FALSE).
  • rena-wasm 0.1.9: a time of 0 in a numeric time column (timesCol) is now read as 0; it was replaced by the row’s position in its conversation.

rENA 0.4.9

New Features

Bug Fixes & Improvements

  • ena.plot.interactive(), ena.export.html(), enaInteractiveOutput() and renderEnaInteractive() now wrap qeviz (>= 0.5.0) instead of a bundled copy of qeviz 0.1.0. Same names and arguments. Means and confidence intervals are now drawn for the groups shown (previously no means were drawn); outlier defaults to FALSE; iqr_factor is deprecated (always 1.5 IQR). The bundled inst/htmlwidgets/ widget is removed.
  • rena-wasm 0.1.8 (@qe-libs/rena-wasm): fit(), accumulate(), tuneWindowSize() and pria() take unitsUsed, the unit keys to model (= units.used in ena.accumulate.data()). Other units’ rows stay in the data as context for the modeled units’ windows, but those units are not modeled: they get no network and play no part in normalization, centring or rotation. Results match ena.accumulate.data(units.used = ...).

rENA 0.4.8

Bug Fixes & Improvements

  • model() now passes rotate_params to the rotation function as its params argument. It previously flattened lists with more than one element, so the two group vectors of a means rotation failed with “$ operator is invalid for atomic vectors”, and a single parameter such as list(x_var = "GameHalf") was silently ignored. This also fixes ona::model(rotate.using = "mean", rotation.params = list(g1, g2)). The nested form list(params = ...) still works.
  • rena-wasm 0.1.7 (@qe-libs/rena-wasm): ordered (ONA) models now match ona::model(), the ONA standard. Node positions use directed least-squares positions, units with no connections are left out of the centring mean but shifted by it, and points and nodes are centred on the origin. This changes the node positions (and, when a dataset has units with no connections, the points) of ONA models built with rena-wasm, such as in the webtool. Unordered (ENA) models are unchanged.
  • rena-wasm 0.1.7: ordered models’ connectionNames and rotation.adjacencyKey now list all n² directed connections in network column order, named as R names them (column j*n + i is "codes[i] & codes[j]", ground codes[i] → response codes[j]). Previously they held the n(n-1)/2 unordered pairs.

rENA 0.4.7

Bug Fixes & Improvements

  • No changes to the R package; this release publishes rena-wasm 0.1.6 (@qe-libs/rena-wasm) with fixes to pria(), which now matches R’s PRIA::pria():
    • The tie-break between candidates that remove the same number of codes now uses the reduced model’s dimension-1 share of the points’ variance across all dimensions (R’s model$variance[1]). It previously used the eigenvalue ratio, which is 0 for means and GMR rotations, so the tie-break never applied to them.
    • Removed codes’ connections are dropped rather than zeroed, as in R, so GMR-rotated candidates are scored correctly.
    • The data are accumulated once rather than once per candidate, making PRIA about 30 times faster (RS.data: 3.4 s to about 0.1 s).
    • weightModel, tensor and the caller’s codeMask are now used, so PRIA scores the same model fit() builds.
    • For ordered (ONA) models, PRIA scores candidates with directed node positions, as R’s ordered pipeline (optimize()) does. PRIA now matches R for ONA and transmodal (TMA) models as well as ENA. Models built by fit() are unchanged.

rENA 0.4.6

Bug Fixes & Improvements

  • No changes to the R package; this release publishes rena-wasm 0.1.5 (@qe-libs/rena-wasm), which missed the 0.4.5 release:
    • Weight models (weightModel: "product", "sqrt", "log") are now applied by libqe’s shared finalize_row_connections kernel (requires @qe-libs/libqe-wasm 0.1.5 or newer), replacing rena-wasm’s own JavaScript copy. Results are unchanged.
    • accumulate() now honours weightModel (it was previously ignored), and tuneWindowSize() carries it through its rebuilds.
    • "log1p" is accepted as an alias of "log".

rENA 0.4.5

New Features

  • accumulate() gains weight_by, the weight model applied to each line’s connection counts before they are summed per unit (the same stage as weight.by in ena.accumulate.data()): "binary", "product" (raw, non-binarized counts), "sqrt" or "log1p" (alias "log"). For ordered networks the weight is applied to each directed cell, and "binary" keeps the raw directed counts. It defaults from binary, so existing calls are unchanged; on RS.data the results match ena.accumulate.data(weight.by = ...) exactly for all four weight models.

Bug Fixes & Improvements

  • ena.ccd() delegates its cross-covariance decay core to libqe’s shared ccd_window kernel.
  • Removed leftover debug output: "Using custom rotation.set." when a custom rotation set is supplied, and "- using custom V matrix" in with.ena.matrix().
  • Removed unused internal code (empty stub functions, an entirely commented-out ena.generate(), unused gmr() backups, and stale Rcpp headers).
  • Updated the R package dependencies to require tma 0.3.4 or newer (for weight_by) and libqe 0.1.3 or newer.

rENA 0.4.4

Bug Fixes & Improvements

  • Restored polynomial trajectory smoothing and animated trajectory plotting against the shared libqe trajectory kernels.
  • Updated the R package dependencies to require tma 0.3.3 or newer and libqe 0.1.2 or newer. rENA now delegates polynomial trajectory fitting for plots to TMA’s trajectory wrapper while TMA delegates the shared numerical kernels to libqe.
  • Separated the legacy ENA trajectory plotting path from the newer ETM-style polynomial curve bridge with internal helper functions and focused documentation.

rENA 0.4.3

Bug Fixes & Improvements

  • Window Size Estimation (ena.tune.window.size):
    • Fixed inaccurate window size estimation by replacing the heuristic SVD stability plateau method with the Cross-Covariance Decay (CCD) algorithm (Shaffer & Cai, 2026). The function’s outward-facing API and usage remain unchanged.
    • Added helper functions ena.ccd.window() and ena.ccd() along with S3 methods (plot.ena.ccd, print.ena.ccd) for direct window estimation and diagnostic curve plotting.
    • Preserved legacy SVD stability behavior via method = "stability".
    • Fixed issue where accumulation call re-evaluation failed when ena.tune.window.size() was called in scripted or nested evaluation environments.
  • Corrected @importFrom stats declarations in ena_space_dist_corr.R.
  • Updated ENAplot documentation and internal variable bindings in zzz.R.

rENA 0.4.2

Improvements

  • Synchronized ENA pipeline and parameter detection with WebAssembly runtime.
  • Undirected code masking index mapping updates.

rENA 0.4.1

Patch

  • $.ena.points and $.ena.matrix return NULL for a column that is not present, matching base $, instead of raising “attempt to select less than one element in get1index”. Callers can now test for an optional column (e.g. ENA_DIRECTION) the ordinary way. Duplicate column names take the first match rather than recursively indexing on a vector.
  • ena.correlations indexes its per-dimension difference matrices by position. They are built with one column per requested dimension, so indexing by the caller’s dimension numbers only worked for the default c(1,2); any other pair errored with “subscript out of bounds”.

rENA 0.4.0

Features

  • Ground-diversity window-size suggestion, see ?ena.gd.window and ?ena.ground.diversity (R/ena.ground.diversity.R). Suggests a sliding window size by locating the peak normalized Shannon entropy of “ground types” (bit-encoded combinations of active codes) across window sizes.

rENA 0.3.1

CRAN release: 2026-02-11

Patch

rENA 0.3.0

CRAN release: 2025-09-29

Features

  • Generalized means rotations, see ?ena.rotate.by.generalized (R/ena.rotate.by.generalized.R)

rENA 0.2.5

rENA 0.2.4

CRAN release: 2022-03-25

Features

  • center.align.to.origin defaults to TRUE
  • Layering in ena.plot.networks controlled by parameter layers, defaults to nodes on top with c("edges", "nodes")
  • Position nodes on a unit circle (either on line from origin or equally spaced) or optimized (still the default)
  • Added export of Microsoft Word document from ena.writeup() => ena.writeup(set, type = 'file')

Bugs

rENA 0.2.3

CRAN release: 2021-04-23

Bugs

  • Removed Vignettes

Feature

  • Added option center.align.to.origin to ena.make.set(), when TRUE centers zero points to the origin (default: FALSE)

rENA 0.2.2.0

Bugs

  • Removed outline from plotted nodes in networks

Features

  • Weighting function now applied at the line level, not at the unit level

rENA 0.2.1.2

CRAN release: 2020-10-11

Bugs

  • Bug fix in window accumulation

rENA 0.2.1.1

CRAN release: 2020-10-02

Bugs

  • Removing test checking for output
  • Including ‘webshot’ as Suggests to fix for issues on CRAN

rENA 0.2.1.0

CRAN release: 2020-09-16

Features

rENA 0.2.0.1

CRAN release: 2019-12-17

  • Fix for updated r-devel to 4.0.0

rENA 0.2.0.0

CRAN release: 2019-10-07

Features

  • New ena() function for easier model generation
  • Updated default model object returned by all methods. See help for ena()
  • Custom S3 methods for removing meta data from data.frames on the ena model.
    • e.g. as.matrix(set$line.weights)

Bugs

  • Fixed bug in accumulation code for forward windows