Visualization¶

The primary schema-2 reporting artifact is AnalysisReport. The CLI writes both analysis.json.gz and a markdown report.md from that object.

For most workflows, start with the generated markdown report. Use Python plotting when you need custom figures or comparisons.

Plot From AnalysisReport¶

AnalysisReport is intentionally simple: metric names map to MetricSummary values. This makes custom plotting straightforward.

import plotly.graph_objects as go

metric = "solv_0[buyables]_rate"
strata = sorted(report.by_stratum)
values = [report.by_stratum[stratum][metric].value for stratum in strata]
ci_low = [report.by_stratum[stratum][metric].ci_low for stratum in strata]
ci_high = [report.by_stratum[stratum][metric].ci_high for stratum in strata]

fig = go.Figure()
fig.add_trace(
    go.Bar(
        x=strata,
        y=values,
        error_y={
            "type": "data",
            "symmetric": False,
            "array": [hi - v if hi is not None else 0 for hi, v in zip(ci_high, values, strict=True)],
            "arrayminus": [v - lo if lo is not None else 0 for v, lo in zip(values, ci_low, strict=True)],
        },
        name="Solv-0[buyables]",
    )
)
fig.show()

Built-In Plot Helpers¶

RetroCast still exposes Plotly helpers for legacy model-statistics objects:

from retrocast.visualization import plot_comparison, plot_diagnostics

fig = plot_diagnostics(stats)
fig.show()

comparison = plot_comparison([stats_a, stats_b], metric_type="Top-K", k=10)
comparison.show()

These helpers are useful for older analysis scripts. For new schema-2 reports, prefer plotting directly from AnalysisReport until the visualization layer is fully rebuilt around the v2 metric names.

Compare Reports¶

The CLI includes one schema-2 comparison command for Pareto-frontier plots:

retrocast compare pareto-frontier compare.yaml --no-open

compare.yaml

benchmark: small
stock: buyables
metric: solv_0[buyables]_rate
output_dir: comparisons
sources:
  - root: data/retrocast
    models:
      - name: model-a
        hourly_cost: 1.0
      - name: model-b
        hourly_cost: 0.5

Use the exact metric key from AnalysisReport.metrics.