feat(dashboard): plot multiple training metrics on one chart (#265) - #304
Open
Paopao-C wants to merge 4 commits into
Open
feat(dashboard): plot multiple training metrics on one chart (#265)#304Paopao-C wants to merge 4 commits into
Paopao-C wants to merge 4 commits into
Conversation
…onAI#265) - Multi-select metrics onto one chart with click-to-toggle legend, hover values, optional dual Y-axes, view-only normalization, and LTTB downsampling. Colorblind-safe Okabe-Ito palette. Frontend-only, reuses the existing /api/jobs/<id>/metric endpoint; no new dependency. - Backward compatible: a single selected metric reproduces today's view. - New CLI export (human-readable + --json) reusing metric_series, with validation that names the failing stage/input. - CPU tests (pytest + node) cover success, malformed input, boundary values, default-unchanged, and a tiny local JSONL fixture. - Docs in docs/cli/observability.rst with input contract, output fields, and copyable examples. Co-Authored-By: Claude <noreply@anthropic.com>
- assignAxes: replace truthy range check with Number.isFinite + compare absolute peak magnitudes so zero-containing ranges ([0, 3]) are not split by a 1e-12 floor (real bug: min==0 was treated as missing). - downsampleLttb: clamp bucket start/end to [1, n-2] and compute the next-bucket average over its actual range instead of two fixed indices; no out-of-bounds. - MetricSelector: add aria-expanded/haspopup/controls + per-item aria-label for keyboard / screen-reader use. - MetricChart hover: name the viewBox-width constant (SVG_WIDTH) and document the pointer-x to step mapping. - metrics_command: guard metric_summaries/metric_series with meaningful errors on malformed dashboard state. - tests: add 7 LTTB/axis edge cases (duplicate steps, NaN/Inf, n near target, uneven gaps, zero-containing ranges, negative-crossing ranges) and delete the extracted helper module at the end of the node run. Co-Authored-By: Claude <noreply@anthropic.com>
- downsampleLttb: skip non-finite candidates explicitly (do not rely on NaN comparison semantics across engines); document the length==target guarantee and the [start, end) half-open bucket convention. - buildMultiMetricPlot: guard empty / all-empty series so Math.min(...) over an empty spread no longer yields Infinity and corrupts coordinates; return []. - assignAxes: unfold the peak-ratio check into a readable if/else. - MultiMetricPlot hoverGuide: draw using the first non-empty plot that has the hovered step, instead of an implicit x=0 fallback. - _load_dashboard_state: document the ARENO_DASHBOARD_ROOT precondition. - tests: +1 regression for all-empty series (24 node checks now). Co-Authored-By: Claude <noreply@anthropic.com>
No behavior change; only formatting (collapse wrapped echo call, wrap the metrics registry tuple, add trailing newline to the test file) requested by the repo's pre-commit hooks.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #265.
What changed
areno metricsexport command (human + --json), guarded state readsUsage
Dashboard: open the metrics panel, toggle metrics in the dropdown — one or
many on one chart. No backend change; reuses
/api/jobs/<id>/metric.CLI:
areno metrics --job --names train/loss,train/reward [--limit N] [--json]
Features
Y-axes (auto when scales differ >1 order of magnitude).
CLI output (human-readable)
Real-data validation
Verified on Kaggle (GPU), Qwen3.5-0.8B run. Multi-metric chart, dual axes,
hover values work on real TensorBoard scalars.
Tests
pytest tests/test_dashboard_metrics_cli_cpu.py— 11 passed (success,malformed input, boundary 1/5000/out-of-range, default-unchanged, fixture).
node tests/test_multi_metric_plot_cpu.mjs— 24 passed (downsample 2000pts, axis assignment, normalization toggles, deterministic, NaN/zero/empty
edge cases).
Notes
dynamic metrics/ responsive resize implemented; responsive is CSS-level,dynamic-metric refresh rides existing poll loop — lighter dedicated test
coverage but exercised via the existing fixtures.