PriorInput

A field for choosing a prior distribution: a family dropdown plus one number input per parameter of the chosen family, in a row of equal-width columns. The model is a PriorSpec whose family and params use numpyro's class and keyword names, so it can be handed straight to a Python model.

Switching family keeps any same-named parameters (loc, scale, …) and fills the rest from the family's defaults.

Examples

R0 prior

Family

loc

scale

{
 "family": "Normal",
 "params": {
  "loc": 0,
  "scale": 1
 }
}

<script setup>
import { ref } from "vue";
const prior = ref({ family: "Normal", params: { loc: 0, scale: 1 } });
</script>

<PriorInput v-model="prior" label="R0 prior" />
<pre>{{ JSON.stringify(prior, null, 1) }}</pre>

Subset of families

Pass families to offer only some of the built-in distributions, in the order given. When mounted without a model, the first allowed family becomes the default.

Scale prior

Family

loc

scale

<PriorInput
  v-model="subset"
  label="Scale prior"
  :families="['Normal', 'LogNormal', 'HalfNormal']"
/>

Preview and summary

preview draws a small density sparkline of the current prior and summary prints it as text, e.g. Gamma(concentration = 2, rate = 1). The sparkline disappears when the parameters do not give a finite density (a zero scale, for instance).

Infectious period

Family

concentration

rate

Gamma(concentration = 2, rate = 1)

<PriorInput v-model="previewed" label="Infectious period" preview summary />

Compact layout

layout="compact" puts the family dropdown and the parameter fields in one row of equal-width columns under the field's label, for a dense parameter panel. Each parameter's name reads as description text beneath its field, and the preview and summary stay below the row. With more than three parameters (TruncatedNormal) the dropdown takes its own row and the parameters share the next one. The demo is as wide as the content of a SidebarLayout sidebar.

R0 prior

Family

loc

scale

LogNormal(loc = 0.3, scale = 0.2)

Reporting delay

Family

loc

scale

low

high

<PriorInput
  v-model="compact"
  label="R0 prior"
  layout="compact"
  preview
  summary
/>
<PriorInput v-model="truncated" label="Reporting delay" layout="compact" />

Custom distributions

Pass your own distributions array to change labels, defaults, steps, or bounds, or to add a family that is not in the preset. The family and every name are whatever your model's library expects (numpyro's class and keyword names for the preset); label is only what people see.

Model prior

Family

<script setup>
import { ref } from "vue";
const custom = ref();
const customDistributions = [
  {
    family: "LogNormal",
    label: "LogNormal (R0)",
    params: [
      { name: "loc", default: 0.3, step: 0.05 },
      { name: "scale", default: 0.2, min: 0, step: 0.05 },
    ],
  },
  {
    family: "Gamma",
    label: "Gamma (generation interval)",
    params: [
      { name: "concentration", label: "shape", default: 4, min: 0, step: 0.5 },
      { name: "rate", default: 1, min: 0, step: 0.1 },
    ],
  },
];
</script>

<PriorInput
  v-model="custom"
  label="Model prior"
  :distributions="customDistributions"
  preview
  summary
/>

Other libraries, parameterizations, and integers

The same descriptors fit a model written against another library, such as a Rust one whose distributions take different names. Two descriptors may share a family to offer alternative parameterizations: the dropdown then appends the parameter names ("Exponential (rate)", "Exponential (scale)") unless you give a label. A loaded value shows the parameterization whose names it carries (descriptorFor does that lookup). A parameter with integer: true is a whole-number field. Libraries that want a flat tagged object instead of { family, params } are one spread away.

Rust prior

Family

scale

Exponential(scale = 2)

{"type":"Exponential","scale":2}

<script setup>
import { ref } from "vue";
const tagged = ref({ family: "Exponential", params: { scale: 2 } });
const taggedDistributions = [
  {
    family: "Exponential",
    params: [{ name: "rate", default: 1, min: 0, step: 0.1 }],
  },
  {
    family: "Exponential",
    params: [{ name: "scale", default: 1, min: 0, step: 0.1 }],
  },
  {
    family: "DiscreteUniform",
    params: [
      { name: "low", default: 1, integer: true },
      { name: "high", default: 6, integer: true },
    ],
  },
  {
    family: "Binomial",
    params: [
      { name: "n", default: 10, min: 0, integer: true },
      { name: "p", default: 0.5, min: 0, max: 1, step: 0.05 },
    ],
  },
];
// {"type":"Exponential","scale":2}, the shape a serde-tagged enum reads.
const toTagged = (spec) => ({ type: spec.family, ...spec.params });
</script>

<PriorInput
  v-model="tagged"
  label="Rust prior"
  :distributions="taggedDistributions"
  summary
/>

The preview sparkline knows the numpyro families' densities only; it stays blank for other families. Vector-valued parameters (a Dirichlet's alpha) are not supported yet.

Python counterpart

cfasim_model.numpyro.to_distribution(spec) builds getattr(numpyro.distributions, spec["family"])(**spec["params"]), so family must be a numpyro class name and each param name one of its constructor keywords; label is the only free text. Strictly positive parameters have min: 0 in the preset, which the number input treats as inclusive: a value of exactly 0 passes here but numpyro rejects it, and the preview sparkline goes blank as a hint.

Accessibility

• The field is a role="group" named by label (or aria-label); inside it, the family SelectBox has a visually hidden "Family" label and every parameter NumberInput a visible one (shown beneath the field in the compact layout).
• The preview is an <svg role="img"> labelled with the summary text.

Model

Name	Type
v-model	PriorSpec

Props

Prop	Type	Required	Default
label	string	No	—
hideLabel	boolean	No	—
ariaLabel	string	No	—
hint	string	No	—
distributions	DistributionDescriptor[]	No	() => NUMPYRO_DISTRIBUTIONS
families	string[]	No	—
preview	boolean	No	—
summary	boolean	No	—
layout	"stacked" | "compact"	No	"stacked"

PriorSpec

interface PriorSpec {
  /** numpyro distribution class name, e.g. "Normal". */
  family: string;
  /** Constructor kwargs for that class. */
  params: Record<string, number>;
}

DistributionDescriptor

One way to parameterize a family. Several descriptors may share a family.

interface DistributionDescriptor {
  /** numpyro class name. */
  family: string;
  /**
   * Visible label and dropdown value. Defaults to `family`, with the parameter
   * names appended when the family appears more than once.
   */
  label?: string;
  params: DistributionParam[];
}

DistributionParam

interface DistributionParam {
  /** numpyro keyword name. */
  name: string;
  /** Visible label; defaults to `name`. */
  label?: string;
  default: number;
  /** Inclusive bounds passed to the NumberInput. */
  min?: number;
  max?: number;
  step?: number;
  /** Whole numbers only: integer field, step 1 unless given. */
  integer?: boolean;
}

descriptorFor

function descriptorFor(
  distributions: DistributionDescriptor[],
  spec: PriorSpec | undefined,
): DistributionDescriptor | undefined;

The descriptor of spec.family whose parameter names best match spec.params, or the family's first one.

NUMPYRO_DISTRIBUTIONS

The default distributions. Every parameter uses step: 0.1; parameters marked > 0 have min: 0.

Family	Parameters (default)
Normal	loc (0), scale (1) > 0
LogNormal	loc (0), scale (1) > 0
HalfNormal	scale (1) > 0
Uniform	low (0), high (1)
Gamma	concentration (2) > 0, rate (1) > 0
Beta	concentration1 (2) > 0, concentration0 (2) > 0
Exponential	rate (1) > 0
TruncatedNormal	loc (0), scale (1) > 0, low (-2), high (2)
Cauchy	loc (0), scale (1) > 0
HalfCauchy	scale (1) > 0
StudentT	df (3) > 0, loc (0), scale (1) > 0
InverseGamma	concentration (3) > 0, rate (1) > 0

priorLabel

function priorLabel(spec: PriorSpec): string;
// priorLabel({ family: "Normal", params: { loc: 0, scale: 1 } })
// => "Normal(loc = 0, scale = 1)"