We recently had a user inquire about how to split their LAS file by classification, resulting in one file with all ground returns (classification = 2) and another with everything else. While we have the `filters.groupby` filter that can do something very close to this, you'd actually end up with individual files for **every** classification present.

Now a brute force solution immediately came to mind. What if we reassigned everything but ground to a common value with `filters.assign` and then used `filters.groupby`?

```json
{
    "pipeline":[
        {
            "type":"filters.assign",
            "assignment":"Classification![2:2]=0"
        },
        {
            "type":"filters.groupby",
            "dimension":"Classification"
        },
        {
            "type":"writers.las",
            "filename":"output_#.las"
        }
    ]
}
```

The obvious drawback to this approach is that while you have succesfully split your point cloud based on the ground/non-ground criteria, you've now lost all classification information for your non-ground points. No good.

We've [briefly discussed](https://github.com/PDAL/PDAL/pull/1693) a new filter that would enable splitting on any number of specified DimRanges, which would've been perfect for this situation, but in the meantime, the following pipeline that creates a temporary dimension first using `filters.ferry` is all we need.

```json
{
    "pipeline":[
        {
            "type":"filters.ferry",
            "dimensions":"Classification=TempClass"
        },
        {
            "type":"filters.assign",
            "assignment":"TempClass![2:2]=0"
        },
        {
            "type":"filters.groupby",
            "dimension":"TempClass"
        },
        {
            "type":"writers.las",
            "filename":"output_#.las"
        }
    ]
}
```

This is of course nearly identical to the first pipeline, except that we have first copied the classification values to a new, temporary dimension called `TempClass`. All assignment and groupby operations are applied to the temporary dimension, leaving the original classifications in tact!