pulsatrix 1.1.0
Loading...
Searching...
No Matches
Visualization

Visualization in two layers. Always built in the core library and unit-tested: versioned JSON documents (json.hpp, viz/document.hpp: pulsatrix.<kind>.v1 for attributions, heatmaps, token relevance, circuit graphs, training logs and feature dashboards), dependency-free SVG figures (viz/svg.hpp, also the pulsatrix_svg tool), and the data transforms (plot_data.hpp/colormap.hpp). Opt-in via PULSATRIX_ENABLE_VIZ: live Dear ImGui + ImPlot widgets and dashboards that read the same data, checked through the demo apps under examples/viz/ rather than GoogleTest, since a live window context isn't testable headlessly. More...

Files

file  json.hpp
 A small strict JSON value, parser and deterministic writer (RFC 8259).
 
file  attribution_bar_chart.hpp
 Horizontal feature-importance bar chart for an Attribution.
 
file  attribution_beeswarm.hpp
 Beeswarm plot of one feature's attribution distribution across many samples.
 
file  attribution_graph.hpp
 Attribution graphs in the JSON that Neuronpedia's graph viewer and circuit-tracer's local viewer read (VIZ-4): nodes on a layer by token-position grid, and weighted links between them.
 
file  attribution_waterfall.hpp
 Cascading waterfall chart from a baseline to a prediction, for local attribution.
 
file  circuit_graph_view.hpp
 Node-link diagram for a CircuitGraph (mechanistic-interpretability visualization).
 
file  colormap.hpp
 Pure, ImGui/ImPlot-independent colormap and normalization functions for viz.
 
file  confidence_meter.hpp
 Length-encoded confidence/uncertainty meter with an always-visible numeric label.
 
file  dataset_statistics_view.hpp
 Per-field distribution histograms + data-quality issue panel for a Dataset.
 
file  document.hpp
 Versioned JSON documents (pulsatrix.<kind>.v1) that every renderer reads.
 
file  explanation_score_card.hpp
 Aggregated per-prediction "what/why/how confident/how trustworthy" panel.
 
file  fonts.hpp
 Fonts for the ImGui windows that cover every script: the default font first, then system fonts merged in as fallbacks, so text in any language and emoji show instead of "?".
 
file  html.hpp
 Self-contained HTML views of the viz documents (VIZ-3): interactive Vega-Lite charts with tooltips, zoom and PNG/SVG export, and plain-HTML text views, in any browser or notebook.
 
file  image_grid_view.hpp
 Scrollable grid of dataset image samples, texture-cached across frames.
 
file  implot_metrics_sink.hpp
 Concrete MetricsSink that buffers training scalars/histograms for live plotting.
 
file  mime_bundle.hpp
 Rich notebook display: mime_bundle_repr() for pulsatrix's types (roadmap NB-1).
 
file  performance_views.hpp
 Figures for the Performance question (KS-3): a reliability diagram of a classifier's calibration, and the error on each slice of the data with its interval and test.
 
file  plot_data.hpp
 Pure Attribution/Dataset -> plot-ready-array transforms (no ImGui/ImPlot include).
 
file  protein_documents.hpp
 Viz documents for proteins (PLM-5): the mutation map, the sequence logo, the contact map and residue tracks, with builders from a protein language model's outputs.
 
file  protein_views.hpp
 Figures and pages for proteins (PLM-5): the mutation map, the sequence logo, the contact map, residue tracks, and a 3D structure colored by any per-residue score.
 
file  report.hpp
 Notebook-format reports from C++: .ipynb and self-contained HTML (roadmap NB-2).
 
file  saliency_heatmap_view.hpp
 Saliency/attribution heatmap using a perceptually uniform colormap.
 
file  series.hpp
 The series document (VIZ-9): measured points with optional intervals, for sweeps and trade-off curves, and its SVG view. The Vega-Lite view is in html.hpp.A series is one line of (x, y) points, each with an optional interval [lo, hi] (a confidence interval, a range over seeds). A panel holds several series on shared axes, and a document holds one or more panels with the same axis labels and scales: small multiples. Either axis can be linear or logarithmic.
 
file  svg.hpp
 Dependency-free SVG figures from viz documents: bar, waterfall, heatmap, token strip, beeswarm, partial dependence, tornado, counterfactual, Morris and Sobol charts.
 
file  text_relevance.hpp
 Token relevance documents from real explanations (VIZ-6a): a tokenizer's Encoding or TOK-4's word scores, turned into a pulsatrix.token_relevance.v1 document whose pieces read as the original text.
 
file  texture_cache.hpp
 Keyed GPU texture cache for uploading decoded image Tensors to OpenGL.
 
file  token_relevance_view.hpp
 The text of a pulsatrix.token_relevance.v1 document with each piece colored by its relevance, as an ImGui widget (VIZ-6a).
 
file  topology_documents.hpp
 The persistence diagram document (TDA-9): the features of each homology dimension, with the certified error bound and any significance results, for the persistence diagram, barcode and Betti curve views in topology_views.hpp.
 
file  topology_views.hpp
 Three views of a persistence diagram document (TDA-9): the persistence diagram, the barcode and the Betti curve.
 
file  training_dashboard.hpp
 Live training dashboard: per-tag run summary plus ImPlotMetricsSink's charts.
 
file  window.hpp
 GLFW + OpenGL3 + Dear ImGui/ImPlot window lifecycle, factored out of every demo app.
 

Detailed Description

Visualization in two layers. Always built in the core library and unit-tested: versioned JSON documents (json.hpp, viz/document.hpp: pulsatrix.<kind>.v1 for attributions, heatmaps, token relevance, circuit graphs, training logs and feature dashboards), dependency-free SVG figures (viz/svg.hpp, also the pulsatrix_svg tool), and the data transforms (plot_data.hpp/colormap.hpp). Opt-in via PULSATRIX_ENABLE_VIZ: live Dear ImGui + ImPlot widgets and dashboards that read the same data, checked through the demo apps under examples/viz/ rather than GoogleTest, since a live window context isn't testable headlessly.