Added heatmap and bounded/normalization to metrics

[Felipedino]

This pull request introduces a new heatmap visualization for model metric comparisons in the results UI, replacing the previous radar chart. It also standardizes the concept of "bounded" metrics (i.e., metrics with a natural [0, 1] range) in both backend and frontend, enabling better color-coding and interpretation of metric values. Several metrics are updated to explicitly declare their boundedness, and the frontend now fetches this metadata to drive the new heatmap's color mapping and annotations.

Visualization improvements:

• Added a new heatmap chart for comparing metrics across runs, replacing the radar chart. The heatmap uses backend-provided bounded and maximize metadata to color-code cells, making it easier to interpret metric performance. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]

• Improved model comparison table: metric and model names are now truncated with tooltips for readability.

Backend metric metadata:

• Introduced a new BOUNDED attribute to BaseMetric and its subclasses, indicating if a metric is naturally in [0, 1]. This is now included in the metric metadata returned to the frontend. [1] [2] [3] [4] [5] [6] [7]

• Updated several metrics to explicitly set BOUNDED, including LogLoss, Bleu, Chrf, and Ter, with appropriate comments on their value ranges. [1] [2] [3] [4]

Metric normalization and documentation:

• Modified LogLoss to normalize its output to [0, 1] by dividing by log(num_classes), and updated its documentation to clarify the meaning of the normalized score. [1] [2]

• Minor documentation formatting fix in cohen_kappa.py for readability.

[Copilot]

Pull request overview

This PR updates the results UI to use a heatmap (instead of a radar chart) for comparing metrics across runs, and introduces backend/FE support for “bounded” metric metadata to drive consistent [0,1]-based color mapping.

Changes:

• Replaced the radar chart option with a heatmap visualization and updated Plotly layout/trace builders accordingly.
• Added BOUNDED to backend metric metadata (BaseMetric.get_metadata) and annotated several metrics with boundedness.
• Improved model comparison table readability by truncating long model/metric names with tooltips.

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
DashAI/front/src/utils/i18n/locales/en/models.json	Adds heatmap label and removes radar label.
DashAI/front/src/utils/i18n/locales/es/models.json	Adds heatmap label and removes radar label.
DashAI/front/src/pages/results/constants/layoutMaking.jsx	Updates Plotly layout behavior for heatmap vs bar.
DashAI/front/src/pages/results/constants/graphsMaking.jsx	Removes radar trace creation; adds heatmapMaking trace builder.
DashAI/front/src/pages/results/components/ResultsGraphsSelection.jsx	Switches chart toggle from radar to heatmap.
DashAI/front/src/pages/results/components/ResultsGraphsPlot.jsx	Renders heatmap traces when selected.
DashAI/front/src/pages/results/components/ResultsGraphs.jsx	Fetches metric metadata and builds the heatmap trace.
DashAI/front/src/components/models/ModelComparisonTable.jsx	Truncates long names with tooltip for readability.
DashAI/back/metrics/base_metric.py	Introduces BOUNDED and exposes it via get_metadata.
DashAI/back/metrics/classification_metric.py	Sets default BOUNDED for classification metrics.
DashAI/back/metrics/classification/log_loss.py	Normalizes log loss by log(num_classes) and marks as bounded.
DashAI/back/metrics/classification/cohen_kappa.py	Doc formatting change for reference URL.
DashAI/back/metrics/translation/bleu.py	Marks BLEU as bounded.
DashAI/back/metrics/translation/chrf.py	Marks CHRF as bounded.
DashAI/back/metrics/translation/ter.py	Marks TER as unbounded.

Comments suppressed due to low confidence (1)

DashAI/front/src/pages/results/constants/graphsMaking.jsx:36

• graphsMaking's public API has changed (it now expects (graphsToView, run, metrics, values, runIndex, theme) and only builds bar traces). There are still call sites in the repo that import this module and call graphsMaking with the previous signature (e.g. front/src/components/pipelines/results/ResultsGraphs.jsx), which will produce incorrect Plotly trace data at runtime. Please either update those call sites in this PR or keep graphsMaking backward-compatible (e.g. via an adapter/overload or a separate export for the pipeline view).

function graphsMaking(graphsToView, run, metrics, values, runIndex, theme) {
  graphsToView.bar = graphsToView.bar || [];

  const colors = getTraceColors(theme);
  const color = colors[runIndex % colors.length];
  const runLabel = run.run_name || run.name || `Run ${runIndex + 1}`;

  graphsToView.bar.push({
    type: "bar",
    name: runLabel,
    x: metrics,
    y: values,
    marker: { color, opacity: 0.85 },
  });

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[cristian-tamblay]

This needs some work as we talked in today's meeting

[Copilot]

Pull request overview

Copilot reviewed 10 out of 10 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (1)

DashAI/back/metrics/classification/log_loss.py:90

• PR description mentions normalizing LogLoss to a bounded [0,1] score, but score() still returns raw sklearn.metrics.log_loss(...) (unbounded, range [0,+∞)). Either implement the normalization (and adjust any metadata like boundedness accordingly) or update the documentation/PR description so the frontend doesn't assume a bounded value.

        Returns
        -------
        float
            Log Loss, lower is better.
        """
        from sklearn.metrics import log_loss

        true_labels, _ = prepare_to_metric(true_labels, probs_pred_labels)
        return log_loss(true_labels, probs_pred_labels)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Felipedino]

Con nico encontramos que los colores están bien.