> ## Documentation Index
> Fetch the complete documentation index at: https://nixtlaverse.nixtla.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Capabilities

> Optional adapter features, declared as mixins and checked when a forecaster is bound

### `SupportsRefit`

Bases: <code>[ArrayModel](#conformalforecast._model.ArrayModel)</code>, <code>[ABC](#abc.ABC)</code>

Adapter can retrain its underlying model on new data.

#### `SupportsRefit.refit`

```python theme={null}
refit(history)
```

Refit the underlying model on `history`.

## Parameters

history : Series, shape (n\_series, T)

### `SupportsQuantiles`

Bases: <code>[ArrayModel](#conformalforecast._model.ArrayModel)</code>, <code>[ABC](#abc.ABC)</code>

Adapter can produce quantile forecasts directly (needed for CQR).

#### `SupportsQuantiles.predict_quantiles`

```python theme={null}
predict_quantiles(history, quantiles)
```

Predict requested quantiles.

## Parameters

history : Series, shape (n\_series, T)
quantiles : NDArray, shape (n\_quantiles,)
Values in (0, 1).

## Returns

Forecast, shape (n\_series, n\_quantiles, horizon)
Quantile forecasts. Note the middle axis indexes quantiles, not samples.

### `SupportsBootstrap`

Bases: <code>[ArrayModel](#conformalforecast._model.ArrayModel)</code>, <code>[ABC](#abc.ABC)</code>

Adapter can produce bootstrap ensemble forecasts (needed for EnbPI).

#### `SupportsBootstrap.bootstrap_predict`

```python theme={null}
bootstrap_predict(history, n_bootstraps, seed=0)
```

Produce `n_bootstraps` predictions from bootstrap resamples of training data.

## Parameters

history : Series, shape (n\_series, T)
n\_bootstraps : int
seed : int, default 0
Never `None`: resampling must be reproducible given the same
fitted state and inputs.

## Returns

Forecast, shape (n\_series, n\_bootstraps, horizon)

### `SupportsCrossValidation`

Bases: <code>[ArrayModel](#conformalforecast._model.ArrayModel)</code>, <code>[ABC](#abc.ABC)</code>

Adapter can run rolling-origin cross-validation natively.

#### `SupportsCrossValidation.cross_validate`

```python theme={null}
cross_validate(n_windows, step_size, refit=False, *, aligned=False)
```

Run rolling-origin cross-validation on the adapter's training data.

## Parameters

n\_windows : int
Number of evaluation windows.
step\_size : int
Number of time steps between successive windows.
refit : bool or int
Whether (or how often) to refit the model between windows.
aligned : bool, default False
`False` returns dense per-series arrays (window axis = k steps
before each series' own end). `True` returns calendar-aligned
cutoffs with NaN cells for late-starting series.

## Returns

predictions : Forecast, shape (n\_series, n\_windows, horizon)
truths : Forecast, shape (n\_series, n\_windows, horizon)

### `SupportsCrossValidationQuantiles`

Bases: <code>[ArrayModel](#conformalforecast._model.ArrayModel)</code>, <code>[ABC](#abc.ABC)</code>

Adapter can produce quantile forecasts via cross-validation.

This is the fast calibration path for CQR (and any future quantile-based
conformal method). Adapters with this capability avoid the per-sample loop
of `predict_quantiles`, instead leveraging the underlying library's
cross-validation API to produce all calibration samples in a single call.

#### `SupportsCrossValidationQuantiles.cross_validate_quantiles`

```python theme={null}
cross_validate_quantiles(n_windows, step_size, quantiles, refit=False)
```

Run rolling-origin cross-validation producing quantile forecasts.

## Parameters

n\_windows : int
Number of evaluation windows.
step\_size : int
Number of time steps between successive windows.
quantiles : NDArray, shape (n\_quantiles,)
Values in (0, 1).
refit : bool or int, default False
Same semantics as :meth:`SupportsCrossValidation.cross_validate`.

## Returns

quantile\_predictions : Forecast, shape (n\_series, n\_windows, horizon, n\_quantiles)
Quantiles are on the **last axis**, matching :class:`QuantileScore`'s
expected input. This differs from :meth:`SupportsQuantiles.predict_quantiles`,
where quantiles are on axis 1 (because `predict_quantiles` produces a
single sample per call). The intentional axis difference avoids an
extra transpose in the CQR fast path.
truths : Forecast, shape (n\_series, n\_windows, horizon)


## Related topics

- [Adapters](/conformalforecast/docs/adapters/index.html.md)
- [Core contracts](/conformalforecast/base.html.md)
- [NeuralForecast](/conformalforecast/docs/adapters/neuralforecast.html.md)
- [Changepoint injection](/synforecast/docs/capabilities/changepoints.html.md)