Skip to contents

Changelog

Source: NEWS.md

rcisignal 0.3.1

Breaking changes

  • save_ci_images() default output now matches rcicr. The default is palette = "grayscale", which writes the raw pixel luminance image (no color palette) that rcicr::generateCI() / rcicr::generateCI2IFC() would produce for the same CI. For each column of signal_matrix, the raw noise is scaled into [0, 1] via the chosen scaling method and averaged with the base via (scaled + base) / 2 before writing. Two new arguments mirror rcicr’s API:
    • scaling = c("independent", "constant", "matched", "none") (default "independent", matching rcicr).
    • scaling_constant = 0.1 (default matches rcicr). Callers that explicitly want the previous colored overlay can pass palette = "diverging"; palette = "fire" continues to produce the unipolar |t|-style yellow-to-red ramp. Code that called save_ci_images(sm, base, dir) and relied on the diverging overlay needs to add palette = "diverging" to preserve the prior output.

Documentation

  • ?save_ci_images and vignette §1.4 rewritten to describe the grayscale default and the rcicr-style scaling / scaling_constant arguments.

rcisignal (development version)

Documentation

  • Group-mean infoVal documentation expanded. New vignette §11 (“Group-level signal summaries”) fully describes the group-mean z statistic (group_mean_z(), surfaced via infoval_report()): how its reference distribution is constructed (one simulated random producer per real producer, averaged before the norm is taken), how it differs from Brinkman et al. (2019)’s per-producer infoVal (different statistic, different null, validated only at the per-producer level), and how to read it appropriately (necessary but not sufficient for interpretability; pair with the per-producer distribution and the spatial agreement map). A side-by-side simulation illustration shows why the matched-N reference distribution shrinks with N – exactly the calibration that makes the group-mean z meaningful. ?infoval_report gains a parallel Details block and an expanded Return description for the group_mean_z_* fields; the printed bullet now points to ?infoval_report for what the z tests. Subsequent section numbers shift by one (old §11 -> §12, §12 -> §13, …, §18 -> §19); all cross-references updated.

  • Dissimilarity inference reframed around the permutation null. ?rel_dissimilarity gains a section explaining why the bootstrap CI on the Euclidean distance is not a test against zero (the L2 norm is biased upward by producer resampling, so its CI excludes zero even for identical conditions) and directing users to null = "permutation" for an above-chance test; the Pearson bootstrap-attenuation bias is now noted too. Vignette §9.3 leads with the permutation null, adds a callout and a worked identical-conditions illustration, and reframes the bootstrap CI as a precision interval; §14.7 reports per-contrast permutation summaries (d, M null, SD null, z, d-ratio, p) and §16 gains a matching caveat. ?run_discriminability documents that its dissimilarity panel is bootstrap-only and points to rel_dissimilarity(null = "permutation") for the test. No change to defaults or computations.

rcisignal 0.3.0

Breaking changes

  • Architecture simplified. The “stage 1 / stage 2” boundary is gone. The rcisignal_group_ci S3 class is dropped; group_ci() now returns a plain numeric matrix (with attr(., "n"), attr(., "img_dims"), attr(., "by_name"), and a lightweight attr(., "ci_level") = "group" hint for save_ci_images()). Stage-1 input guards (the 14+ abort_if_group_ci() calls across reliability, discriminability, and infoVal functions) are removed. The _pkgdown.yml “Stage 2 - Group-averaged CIs” reference section is replaced with “Compare multiple CIs”; correlogram, distance matrix, and MDS plot functions are no longer mis-classified as group-only.
  • Removed compute_infoval_summary() (no deprecation alias). infoval_report() (the renamed diagnose_infoval(); see below) covers the same use case and supports both 2IFC and Brief-RC natively. check_response_inversion() and check_rt_infoval_consistency() no longer take rdata/stimuli as their only noise source: both now accept noise_matrix = for the Brief-RC path and use the package’s native infoval() pipeline internally, so Brief-RC is fully supported (previously these checks returned "skip" for Brief-RC).
  • Renamed diagnose_infoval() to infoval_report() (no deprecation alias). The new name better reflects what the function actually does: it produces a per-producer infoVal summary together with three sanity checks (masked-vs-unmasked z lift, group-mean z against a matched reference, random-responder calibration). The “diagnose” framing implied pathology-hunting, but the function is the canonical entry point for any per- producer infoVal report, healthy data included. Migration: rename diagnose_infoval(...) -> infoval_report(...). Same arguments, same return value (still an rcisignal_diag_result). Inside the run_diagnostics() output, the result is now at report$results$infoval_report (was report$results$diagnose_infoval).

New features

  • ci_from_responses_briefrc() and ci_from_responses_2ifc() gain a group_by = argument. When supplied, the return list additionally contains $group_ci, the per-group matrix built by group_ci(signal_matrix, responses, by = group_by, col_participant = col_participant) internally. The per-producer $signal_matrix is also returned. Pass either or both to downstream functions.
  • New export save_ci_images(). Writes rendered CIs to PNG or JPEG, one file per column of a signal_matrix. Works for both per-producer (filename prefix ind_ci_) and group-level (prefix group_ci_) matrices; prefix is auto-derived from the new ci_level attribute and can be overridden. Supports both the "diverging" and "fire" palettes via the same compositing path as plot_ci_overlay() and plot_agreement_map().
  • plot_agreement_map() gains a bar_label = argument for customizable colorbar labels. Defaults to "t" for the diverging palette and "|t|" for the fire palette. The new plot.rcisignal_rel_agreement_map_test() method threads the argument through. Use e.g. bar_label = "Degree of agreement (|t|)" for non-technical audiences.

Documentation

  • README: the “Showcase: Oliveira et al. (2019)” section is renamed to “Using rcisignal on a real data set” and now leads with a 2x2 fire-palette agreement-map figure. The caption explains that brighter pixels mark image regions where producers in the RC task agree more strongly on the location of the target representation in their perception.
  • Vignette: the “two-stage pattern” framing is removed. group_ci() is reintroduced as a convenience helper rather than an architectural boundary. §1.3 gains a save_ci_images() example block showing exports for both per-producer and group CIs. compute_infoval_summary() references are rewritten as infoval_report() calls.

rcisignal 0.2.0

Breaking changes

  • group_ci() redesigned. The vector / list-of-vectors by argument is gone. New signature:

    group_ci(signal_matrix, responses, by,
         col_participant = "participant_id", drop = TRUE)

    responses is the same trial-level data frame you handed to ci_from_responses_*(); by is the name of the grouping column (or a character vector of column names for factorial grouping). Producer-to-group alignment happens internally via colnames(signal_matrix). Migration:

    # Before:
cond <- responses$condition[match(colnames(sm), responses$participant_id)]
gcis <- group_ci(sm, by = cond)

# Now:
gcis <- group_ci(sm, responses, by = "condition")

    group_ci() also validates that every producer in colnames(signal_matrix) is present in responses, and that each producer’s by column(s) are constant across their rows. Both checks fail with teaching messages naming the offending producer.

  • shift_mask() arguments renamed and vertical sign flipped. down -> vertical, right -> horizontal. The new vertical follows the math / y-axis-up idiom (positive moves the mask up, negative moves it down); horizontal keeps the intuitive convention (positive right, negative left). Migration: rename down = N -> vertical = -N (note the sign flip), right = N -> horizontal = N. No deprecation alias.

  • Column-name arguments standardized on the col_* convention in the two CI builders. Renamed: ci_from_responses_briefrc(participant_col, stimulus_col, response_col) -> (col_participant, col_stimulus, col_response); same for ci_from_responses_2ifc(). Every other responses-consuming function (check_*, diagnose_*, compute_*, run_*) already used col_*; the package is now uniform. No deprecation alias; calls using the old names fail with the standard R unused-argument error.

New features

  • infoval() gains a responses = NULL, col_participant = "participant_id" path. When trial_counts is omitted, infoval() derives it internally via table(responses[[col_participant]]) and matches against colnames(signal_matrix). Existing scripts that pass trial_counts directly continue to work unchanged.

  • run_reliability() validates noise_matrix at function entry when null = "random_responders", with a teaching message pointing at read_noise_matrix(). Replaces a deeper cryptic abort that used to fire inside rel_split_half().

Documentation

  • Vignette §1.3 (two-stage pattern), §10.5, and the worked-example cross-references updated to use the new group_ci() idiom. §9 (infoval) and roxygen @examples on infoval() updated to lead with the responses path.

rcisignal 0.1.9

Breaking change

  • ci_from_responses_2ifc() argument baseimage renamed to base_image to match ci_from_responses_briefrc(). The new argument now accepts the same three input forms as Brief-RC: a numeric matrix in [0, 1], a path to a PNG / JPEG, or a string label naming an entry in the rdata’s base_faces list (the historical 2IFC form). Matrix and path inputs are injected into a fresh temporary copy of the rdata under a synthetic label before the call into rcicr. No deprecation alias; calls using baseimage = "..." will fail with the standard R unused-argument error.

  • The same rename cascades through the five downstream 2IFC diagnostic functions and the noise reader: diagnose_infoval(), compute_infoval_summary(), check_rt_infoval_consistency(), check_response_inversion(), run_diagnostics(), plus read_noise_matrix(base_image = ...). All accept the same three forms via the same internal resolver.

Documentation

  • Tutorial vignette cleaned up for readability per a user-driven audit:
    • All user-facing prose (vignette, README, NEWS, roxygen, cli message bodies) switched from British to American English. Argument names and match.arg() value strings (e.g., centre, colour_bar, method = "normalised") were left in place to preserve the documented API surface.
    • DESCRIPTION Language field updated en-GB -> en-US.
    • plot_ci_correlogram(), plot_ci_distance_matrix(), and plot_ci_mds() @param cis and the worked-example chunks in vignette §10.5 / §10.6 / §10.7 now lead with the named- matrix input form (one column per CI, built outside the call with cbind(name = rowMeans(...))). The named-list and group_ci() forms are still accepted; they are documented as the back-compatibility / convenience paths. Mixed group-CI + individual-producer correlogram example added.
    • §6.1 (compute CIs) now cross-links to §1.3 (two-stage pattern) and group_ci() as the canonical group-CI route.
    • Removed developer-facing asides that leaked into the user-guide register (the “eval = FALSE for faster vignette rendering” note in §12.0; the precompute-null reuse aside in §7.1; the attr(res$signal_matrix, "source") and attr(signal, "source") example lines in §6.1 and §6.2).

rcisignal 0.1.8

New features

  • group_ci() (new exported function). Stage-2 aggregator that collapses a per-producer signal_matrix (pixels x n_producers, as returned by ci_from_responses_briefrc() / ci_from_responses_2ifc()) into a per-group matrix (pixels x n_groups) suitable for the existing distance-matrix, MDS, and correlogram plot functions. Accepts an atomic grouping vector or a named list of vectors (factorial grouping). Returns a numeric matrix classed c("rcisignal_group_ci", "matrix", "array") with per-group producer counts in attr(., "n") and img_dims inherited from the input. Replaces the hand-rolled rowMeans(signal_matrix[, group_idx]) users were writing.

  • New S3 class rcisignal_group_ci with print() and as.list() methods. The print method surfaces per-group producer counts and reminds the reader that per-producer information has been averaged out.

Robustness

Documentation

  • New “Two-stage pattern” section in vignette("rcisignal"), placed immediately after the Overview. The package now names the two stages: stage 1 (per-producer signal_matrix, produced by ci_from_responses_*(), consumed by every reliability / discriminability / infoVal function) and stage 2 (group-averaged matrix, produced by group_ci(), consumed by the distance-matrix / MDS / correlogram plots).

  • pkgdown reference index regrouped to mirror the two stages. Visitors to the package website now see “Stage 1 — per-producer CIs” and “Stage 2, group CIs” as the first two reference sections.

  • README gains a small two-stage diagram showing how responses

    • noise_matrix + base_image flow through stage 1 into the reliability / infoVal / discriminability surfaces, and optionally into stage 2 via group_ci() for RDM / MDS / correlogram analyses.

  • @seealso on ci_from_responses_briefrc() and ci_from_responses_2ifc() now points at group_ci() as the optional stage-2 follow-on.

rcisignal 0.1.7

New features

  • plot_agreement_map() gains base_image and alpha_max arguments. When base_image is supplied (numeric matrix or PNG/JPEG path), the t-map is composited on top of the grayscale base; out-of-mask and subthreshold pixels render fully transparent, and the per-pixel opacity scales |t| / zlim_max up to alpha_max. Works for both palette = "diverging" and palette = "fire". With base_image = NULL (default) the existing flat-panel rendering is unchanged.

  • plot.rcisignal_rel_cluster_test() gains the same base_image and alpha_max arguments, so the between-condition t-map (or TFCE map) can be rendered with anatomical context in one call. Diverging palette only; the cluster t-map is intrinsically signed and is not given a |t|-only view.

  • plot.rcisignal_rel_pairwise_report() gains base_image and alpha_max, threaded through to each per-pair cluster-test plot. When base_image is a path string it is read once and reused for every panel.

  • plot.rcisignal_rel_agreement_map_test() (new S3 method). plot() of an agreement_map_test() result now renders the observed t-map directly, with the FWE-significant pixel boundary outlined in black by default. Accepts palette, threshold, zlim, base_image, alpha_max, and a show_contour toggle. To support this, the agreement_map_test() result gains an $img_dims field (additive; existing fields unchanged).

Documentation

Internal

  • Private helpers resolve_base_for_overlay() and composite_rgb_over_gray() factored into R/utils.R so plot_ci_overlay(), plot_agreement_map(), and the cluster-test plot method share one base-loading and compositing path. No user-visible signature change.

rcisignal 0.1.6

New features

  • plot_agreement_map() gains a unipolar palette = "fire" option that displays |t| on a single-hue (pale yellow to deep red) ramp. Use when the question is “where do producers have a consistent opinion” and direction is not needed; the "fire" view discards sign by design. The default palette = "diverging" is unchanged in API; see the color-direction fix below.

  • plot_ci_correlogram() (new exported function). Renders a publication-ready Pearson-r matrix across multiple group-mean CIs. Inputs: a named list of CIs (vectors or per-producer signal matrices; group means computed automatically). Options: full / upper / lower triangle, optional face / upper-face / lower-face masking via make_face_mask(), three diverging palettes (RdBu / PuOr / BrBG), direct save to PNG (600 dpi) or PDF via the file = argument with extension auto-detection. The §12.6 worked-example figure is now produced by this function.

  • plot_ci_distance_matrix() (new exported function). All-vs-all pairwise Euclidean distance matrix across a named list of group-mean CIs. Same beginner-friendly input format as plot_ci_correlogram(); the magnitude metric recommended by §8.3 instead of Pearson r. Supports raw (method = "raw", default) or resolution-normalised (method = "normalised") distance, optional face / upper-face / lower-face masking, four sequential colorblind-safe palettes (viridis / inferno / plasma / rocket), full / upper / lower triangle rendering, and direct PNG (600 dpi) / PDF save.

  • plot_ci_mds() (new exported function). Classical multidimensional-scaling projection of a named list of group-mean CIs (stats::cmdscale()). Auto-selects the smallest dimensionality whose Kruskal stress-1 against the original Euclidean distances reaches the “good” band (default threshold 0.05, Kruskal 1964); renders a grid of all choose(k, 2) pairwise dimension panels when the auto-selected k > 2. Users can force a specific k (e.g., k = 2L for a single paper-figure panel). Returns an S3 object (rcisignal_mds) with a one-screen print() method; $mds_points exposes the n_cis x k_selected coordinate matrix in the Euclidean MDS space, $stress_by_k and $variance_pct_by_k expose the full dimensionality-selection trace. Supports optional categorical groups (point color) and shapes (point pch) arguments for grouped scatters in multi-condition designs. Same masking and PNG/PDF save options as the other plot helpers.

  • New private helper prepare_ci_matrix() (internal, not exported) centralises the named-list-of-CIs validation and per-producer reduction across plot_ci_correlogram(), plot_ci_distance_matrix(), and plot_ci_mds(). No user-visible behavior change to plot_ci_correlogram().

Behavioral change

  • Color direction fixed on the diverging-palette plots that use hcl.colors("RdBu", ...). plot_agreement_map(palette = "diverging") and plot.rcisignal_rel_cluster_test() previously rendered positive values as red and negative as blue, opposite to what their help pages and to plot_ci_overlay() documented. Both now render positive = blue, negative = red, matching every other diverging plot in the package and matching the long-standing help-page docs. Visual effect: rendered figures from prior versions of these two functions look like their colors have been swapped. The new plot_ci_correlogram() follows the same convention from the start.
  • The worked-example correlogram figure (vignettes/figures/oliveira_2019/trait_ci_correlogram.png) is regenerated to match the corrected convention. Same data, flipped colors.

Breaking change

  • plot_agreement_map(palette = "viridis") is removed. The previous viridis branch plotted signed t against a palette with no neutral midpoint, so zero rendered at a perceptually arbitrary color. The new "fire" option supersedes that use case correctly. The argument is now palette = c("diverging", "fire").

Documentation

  • plot_agreement_map() help page rewritten to be explicit that on the default diverging palette both deep red and deep blue indicate strong agreement (sign in hue, magnitude in saturation); “no agreement” is the neutral color (white), not red. The new "fire" palette is documented as a sign-discarding magnitude-only view with explicit pointers to recover direction via the diverging palette or plot_ci_overlay().
  • Vignette §10.2 gains a one-paragraph aside introducing the fire option and reiterating the two-channel reading of the diverging palette.
  • rel_dissimilarity(): the Pearson correlation fields ($correlation, $boot_cor, $ci_cor, $boot_se_cor) are reframed as a secondary summary instead of being slated for removal. They stay in the API. The help page, print() output, plot panel title (“Pearson r (secondary)”), and vignette §8.3 now explain (a) why Euclidean distance is the recommended primary statistic (image-domain scaffolding gives r a positive chance baseline that does not cleanly mean “similar”), and (b) how to use r carefully if it must be reported (relative comparisons across pairs against a permutation null, not absolute values against zero). The prior “will be removed in v0.2.0” wording is withdrawn.
  • Vignette: editorial sweep of dev-facing prose. Replaced “Loaded from cache, … on this dataset:” lead-ins (five occurrences) with neutral “On this dataset, … :” framing. Removed roadmap aside about a possible future Rcpp accelerator for simulate_*_data(). Condensed the §4.2 list of rdata bookkeeping fields. Replaced two “self-critical note” author asides in §8.2 and §8.3 with neutral “one caveat” framing. Dropped a v1.0.x version label in §12.2. Rephrased the §14 pointer at NEWS.md as news(package = "rcisignal").

Internal

  • plot_agreement_map() invisibly returns two new fields: zlim (the color scale used) and palette (the palette name). Backwards compatible only in the sense that no caller was relying on the prior list having exactly four names; tests that asserted on names(...) were updated.

rcisignal 0.1.5

Behavioral change

  • plot_mask_overlay() now always renders a single panel (the base image with the mask overlaid). The side_by_side = argument is removed (the base-alongside-base layout duplicated the base image and added nothing beyond what the single overlay panel already shows). The plot region is now forced square (pty = "s") to avoid the empty canvas that appeared around the image on wide / tall devices.

New features

  • New shift_mask() exported helper. Slides a logical face-region mask by down and right pixels and returns the shifted mask in the same input shape (matrix or column-major vector). This was previously a copy-paste recipe in vignette §4.5; it is now a real API surface with help page and tests. Useful for fine-tuning the elliptical make_face_mask() regions on a non-default base; for the rectangle regions, prefer the region_bounds argument.
  • New region_bounds_from_pixels() exported helper. Converts pixel-coordinate rectangle bounds (c(row_min, row_max, col_min, col_max) in 1-indexed image pixels) to the 0-1 image fractions that make_face_mask(region_bounds = ...) expects. Bridges visual inspection of a base image (where users think in pixels) and the parametric API (which expects fractions).
  • plot_mask_overlay() and plot_face_mask() gain a region = shortcut: plot_mask_overlay(base, region = "left_eye") builds the mask internally via make_face_mask() instead of forcing the caller to construct it separately. A region_bounds = pass-through arg is available for tuning the rectangle regions in the same call.

Tests

Documentation

  • plot_ci_overlay() second example switched from signal_region = "eyes" to signal_region = "left_eye" to surface the new rectangle eye regions on a frequently-read help page (the right-eye rectangle and the rest of the face stay untouched, illustrating regional independence).
  • Vignette §4.5 mentions of plot_mask_overlay() reworded for the single-panel design (“overlay on the base image” rather than “base alongside base + mask”).
  • Vignette §4.5 mask-tuning section rewritten to use the new exported shift_mask() directly instead of redefining it inline (the previous copy-paste recipe is gone). The region_bounds tuning subsection now demos region_bounds_from_pixels() for the common case where the rectangle’s edges are eyeballed in pixel coordinates against a zoomed-in base image, and points at the plot_mask_overlay(region = ..., region_bounds = ...) shortcut.

Tooling

  • .gitignore now excludes tests/testthat/data. rcicr’s computeInfoVal2IFC() uses bare write(...) for status strings, which writes to a file literally named data (R’s default destination for single-argument write). The new simulate-diagnostic-chain regression test triggers this path and would otherwise leave a dirty working tree after every test run.

rcisignal 0.1.4

Behavioral change

  • make_face_mask(region = "eyes") now returns a single wide axis-aligned rectangle covering both eyes ear-to-ear and from the eyebrows down to just below the eye line, instead of the two small ellipses used in v0.1.0 - v0.1.3. The new rectangle is independent of the full-face oval geometry (centre, half_width, half_height); rectangle bounds are set directly via the new region_bounds argument. This makes per-eye-line tuning straightforward and removes the previous “both eyes move together” coupling that made manual alignment to a specific base image awkward. Reverse-correlation results computed against the “eyes” region in earlier versions will shift numerically because the masked pixel set is broader; the Oliveira-2019 worked-example tables in vignette §12 were refreshed with the new geometry.

New features

  • New region = "left_eye" and region = "right_eye" values for make_face_mask(). Each returns an axis-aligned rectangle around the viewer’s left or right eye, independent of the other eye and of the full-face oval. Bounds default to a heuristic centered-face geometry and can be overridden via region_bounds.
  • New region_bounds = NULL argument on make_face_mask(). Pass a length-4 numeric vector c(row_min, row_max, col_min, col_max) in 0-1 image fractions to override the default bounds for any of the three rectangle regions ("eyes", "left_eye", "right_eye"). Errors if supplied for an elliptical region.

Documentation

  • Vignette §4.5 (Face-region masks) rewritten to introduce the eight built-in regions (five ellipses + three rectangles) and the two tuning routes: region_bounds for rectangles vs centre / shift_mask() for ellipses. The manual-positioning example that used to slide both eye ovals via shift_mask() has been replaced by two examples: a shift_mask() demo on the mouth ellipse (where the slide workaround is still the right tool) and a region_bounds demo on the left_eye rectangle (showing the new precision-tuning knob).
  • Vignette face-mask figures regenerated to include left_eye and right_eye panels and the new mouth_demo_* / left_eye_demo_* tuning panels. The old artificial_eyes_demo_* files were removed.
  • README infoval() per-region table refreshed: the “Eyes” column now reflects the wider rectangle geometry (computed on the same Oliveira-2019 Study 1 data, 1000 reference draws, matched trial counts). The other columns are unchanged.
  • Citation-accuracy sweep: removed several incorrect attributions of make_face_mask() region geometries to Schmitz, Rougier, & Yzerbyt (2024) across the vignette, README pointer, and make_face_mask() roxygen. Schmitz et al. (2024) used an oval mask for infoval() calculation but did not specify the oval’s parameters and did not define any sub-region (eyes / nose / mouth / upper-face / lower-face) geometry. The package’s region defaults are now correctly described as this package’s own heuristics; the prior practice of applying an oval before pixel-wise CI metrics is credited to Oliveira et al. (2019), Ratner et al. (2014), and Schmitz et al. (2024) jointly. Ratner et al. (2014) added to the vignette bibliography. Schmitz et al. (2024) is still cited where it applies (Brief-RC structure, genMask() formula, Brief-RC 12 / 20 variants).

rcisignal 0.1.3

New features

  • plot_mask_overlay() lets you sanity-check that a face mask aligns with a specific base image before running CI / reliability / discriminability analyses. Accepts a base image (numeric matrix or PNG/JPEG path) plus a mask in any of the package’s supported forms (logical/numeric vector, matrix, or PNG/JPEG path) and draws either a side-by-side base + overlay pair or just the overlay panel.
  • plot() is now defined for rcisignal_rel_pairwise_report objects returned by run_discriminability_pairwise(). One call lays out one cluster t-map per pair in a square-ish grid (same blue/red sign convention and FWE contours as the per-pair plot.rcisignal_rel_cluster_test()). A warning fires above max_pairs = 12; pass max_pairs = Inf to silence.
  • plot() on a rcisignal_rel_cluster_test result gains a color_bar = TRUE argument (default preserves prior behavior). Set FALSE to suppress the per-panel color bar when packing many panels into a small grid; the new pairwise plot method uses this internally.

Documentation

  • Vignette §8.4 and §8.5 now show plot(rep) as the one-call view for the two-condition and pairwise discriminability orchestrators, with per-panel plot(rep$results$...) shown as the custom-title alternative. The run_discriminability_pairwise() help page is updated in lockstep.

rcisignal 0.1.2

New features

  • simulate_2ifc_data() and simulate_briefrc_data() gain an rdata_dir argument and now return a self-contained $stimuli list. The simulation object survives saveRDS()/readRDS() and knitr cache = TRUE across R sessions: pass an explicit rdata_dir to keep the stimuli .Rdata at a stable path, or hand $stimuli to downstream consumers in place of $rdata_path.
  • ci_from_responses_2ifc(), diagnose_infoval(), compute_infoval_summary(), check_response_inversion(), check_rt_infoval_consistency(), and run_diagnostics() gain a stimuli = argument as an in-memory alternative to rdata_path = / rdata =. When both are supplied stimuli takes precedence and a warning is emitted.
  • ci_from_responses_briefrc() argument base_image_path is renamed base_image and now accepts either a numeric matrix in [0, 1] (e.g. sim$base_face) or a file path. The argument is now optional when scaling = "none" (the default), since the base face only feeds the visualization-only $rendered_ci field. The old base_image_path keeps working for one release with a deprecation warning.
  • simulate_2ifc_data() and simulate_briefrc_data() now also write the base face as a standalone PNG alongside the stimuli .Rdata (rcisignal_sim_2ifc_base_face.png / rcisignal_sim_briefrc_base_face.png under rdata_dir) and expose its path on $base_image_path.

Fixes

  • simulate_briefrc_data() now also writes an rcicr-format stimuli .Rdata (stable filename rcisignal_sim_briefrc_stimuli.Rdata under rdata_dir) for symmetry with the 2IFC path. The Brief-RC consumers still read $noise_matrix directly so the file is informational rather than required.

Behavioral change

  • plot_ci_overlay() now uses positive = blue, negative = red (matching plot_agreement_map() and the cluster-test plots). Previously it rendered positive = red, negative = blue, which flipped the color reading between the package’s three diverging plots. Saved overlay PNGs from prior versions will look mirrored if regenerated. The four worked-example pairwise figures in vignettes/figures/oliveira_2019/ and the README captions were refreshed to match.

Documentation

rcisignal 0.1.1

New features

  • ci_from_responses_briefrc() now accepts method = "briefrc20" in addition to "briefrc12". Both Brief-RC variants are validated in Schmitz, Rougier, & Yzerbyt (2024). The CI computation is identical for both (the genMask() formula is symmetric in the per-trial split); the method argument is recorded as $method in the result list as provenance metadata.

Documentation

  • Vignette §13.1 rewritten to document Brief-RC 12 and Brief-RC 20 as the two validated variants. Removed the inaccurate reference to a Brief-RC 6 variant (Schmitz et al. mention 4 / 6 / 8 / 10 only as future research directions, not as published variants).
  • Vignette §15 paragraph 5 (group-mean infoVal interpretation) rewritten to remove an unsourced “5-10x per-producer median” claim and to flag that the sqrt(N)-style inflation of group-mean z is conditional on producer-level signal alignment, not structural. Brinkman et al. (2019) numerical claims and recommendations are retained with explicit page citations.
  • Vignette §15 paragraph 1 reframed to drop unsourced “10-30% / 70-90%” pixel-fraction figures while keeping the Frobenius-norm dilution mechanism.
  • README validation section trimmed to a brief pointer; full validated-vs-unvalidated breakdown now lives in vignette §1.2.
  • New educational background paragraphs added to §7 (split-half / Spearman-Brown, ICC), §8 (multiple-comparisons problem, cluster permutation, k-connectivity, TFCE, Frobenius/Euclidean dissimilarity), and §9 (per-producer and group-mean infoVal).

rcisignal 0.1.0

First release of the package. Provides a consolidated toolkit for quality assessment of reverse-correlation data and classification images: input-side diagnostics (response coding, RT, alignment, balance) and output-side reliability (pixel-level reliability, cluster inference, per-producer infoVal).