Mostly for rejection of outlying flux values during spectral fitting, the DAP implements a boxcar filtering class that allows for masked pixels and a local calculation of the rms. Although not as fast as other filtering algorithms (e.g., `scipy.ndimage`), these other routines (as far as I know) do not allow for exclusion of masked pixels.

Here, we provide an example use of the function that demonstrates the accuracy of the spectral resolution matching. This is pulled from a unit test in test_filter.py.

First, we build a fake spectrum with noise and a set of outliers. The seed for the noise is fixed so that the test is deterministic.

```import numpy

rng = numpy.random.default_rng(1001)
wave = numpy.arange(3650.0, 10501.0, 0.5)
x = (wave - numpy.mean(wave))/(wave[-1]-wave[0])
sig = 1/(0.5 + 0.5*(x+0.5)**2)
flux = 3 + 0.1*x + 0.5*x**2 + 3*(x+1)**4
dflux = rng.normal(scale=sig)
indx = rng.integers(x.size, size=100)
dflux[indx] += rng.normal(scale=10*sig)[indx]
```

Then I mask a few regions just for demonstration:

```indx = ((wave > 4567) & (wave < 4587)) \
| ((wave > 5567) & (wave < 5587)) \
| ((wave > 9067) & (wave < 9087))
```

Finally I perform the filtering; see `mangadap.util.filter.BoxcarFilter`. Note that the class currently expects to perform the filtering on multiple vectors simultaneously such that its attributes are two dimensional, even in this 1D example. The result of the filtering and the isolated outliers are shown in the plot.

```bf = BoxcarFilter(100, lo=3., hi=3., niter=None, y=sim, local_sigma=True)

from matplotlib import pyplot
pyplot.plot(wave, sim, zorder=1)
pyplot.plot(wave, bf.smoothed_y[0], zorder=2)