Linked distributions and intervals¶

Experimental: the API may change.

Connect empirical distributions to individually selectable observations and supplied uncertainty intervals. ECDFView and IntervalView are development APIs under inklet.experimental.browser; they remain opt-in experimental APIs.

Open the interactive figure · Complete Python recipe

The example contains 20 simulated batches from two processes. Process A has one missing estimate, so its distribution uses nine observations; Process B uses ten. A batch with missing bounds retains its center marker. Orange marks link the selected batch across panels. All values are invented MIT material by Mark Marosi. The supplied ranges are illustrative, not confidence intervals.

Empirical distributions¶

from inklet.experimental.browser import BrowserFigure, ECDFView
from inklet.selection import KeyedTable

table = KeyedTable("observations", {
    "id": ["a", "b", "c", "d"],
    "value": [12, 14, 14, None],
})
figure = BrowserFigure(table, [
    ECDFView("distribution", "value", (5, 25), x_label="Cycle time / s"),
])

The empirical cumulative distribution is count(value <= x) / n, where n counts nonmissing source observations in this panel. Tied values share the same fraction: both rows at 14 above lie at 1.0. This follows the standard ECDF definition. It does not assume a parametric distribution or estimate a density.

The gray staircase is a fixed reference distribution. Filtering hides observation markers but leaves that curve and its denominator unchanged, even when no rows are visible. This compares a selection against an explicit source population. To change that population, compile a revision or use replace_data().

Only markers are selectable. The reference curve is not pickable. Tied markers overlap: pointer picking selects the last source-painted row, while the data table lets you select either row. Source order and IDs remain unchanged even though calculating the staircase sorts values.

Tooltips and the accessible data table report each observation's cumulative fraction. A separate derived column is shown for each ECDF panel; an em dash means that row has no fraction in that panel. Source columns are preserved.

Values and the x domain must be numeric. None is reported and excluded from the denominator. An empty population draws no curve or markers but retains its axes. The fixed y domain (0, 1.05) gives markers at probability 1 room above them. Domains and reference populations stay fixed during page pan/zoom.

Supplied intervals¶

from inklet.experimental.browser import BrowserFigure, IntervalView
from inklet.selection import KeyedTable

table = KeyedTable("estimates", {
    "id": ["batch-1", "batch-2"],
    "batch": [1, 2], "estimate": [12, 14],
    "lower": [10, None], "upper": [15, None],
})
figure = BrowserFigure(table, [
    IntervalView("ranges", "batch", "estimate", (0, 3), (5, 20),
                 lower="lower", upper="upper",
                 interval_label="Illustrative supplied range; not a confidence interval",
                 x_label="Batch", y_label="Cycle time / s"),
])

Bounds are absolute endpoints, not distances from the estimate. Complete intervals must satisfy lower <= center <= upper. Reversed bounds are rejected even with a missing center. Missing x/y coordinates omit the row's marks; a missing bound retains the center and omits the interval.

interval_label is required to identify the method or meaning. Inklet draws supplied values; it does not calculate confidence levels, bootstrap estimates or significance. Filtering never recalculates bounds. For actual uncertainty results, include the method, level and source population in this label and your external caption.

Use orientation="horizontal" for a numeric x value axis and y position axis. The position axis may be a TimeAxis; the value axis must be numeric. Both directions support reversed domains. Caps, markers and strokes use millimetres: cap_width_mm=2, radius_mm=0.6, line_width_mm=0.3. Selection modestly thickens interval strokes, and every piece selects the same row.

Facets and diagnostics¶

Both views support FacetView. Each ECDF facet has its own source population and denominator; empty categories retain their panels. The example uses shared physical margins and common domains for comparison.

Expand Statistical methods and populations in the browser to inspect the method and population. This report updates on revision switches. In Python, read figure.payload()["layers"][n]["statistics"]:

missing_intervals counts drawable centers missing one or both bounds. The ordinary layer missing count describes rows without x/y coordinates. Reports describe the compiled source data, independently of the visibility filter.

Save, revise and export¶

python examples/v4/statistical_views.py --output out/v4-statistics
python examples/v4/statistical_views.py --state /path/to/view.json --output out/v4-restored
python examples/v4/statistical_views.py --revised --state /path/to/revised-view.json --output out/v4-revised

The recipe writes offline HTML, SVG and JSON state. Its revision corrects Process A's first estimate and removes Process B's tenth batch. Reference curves and intervals rebuild together. Removed selected/filtered IDs require an explicit drop policy; see data replacement.

Weighted or censored distributions, recomputed filtered summaries, histogram bin selection, density estimation and inference are outside this increment. Existing static histograms, boxplots and violins remain available through the ordinary plotting API.