Dtypes in ML

[mmann1123]

Resolved in PR #87

Code Current in conversion.py

def polygon_to_array(
        self,
        polygon,
        col=None,
        data=None,
        ...
        dtype="uint16",
        ):
 ....

 if col:
            shapes = (
                (geom, value) for geom, value in zip(dataframe.geometry, dataframe[col])
                )

 else:
            shapes = dataframe.geometry.values

 varray = rasterize(
                    shapes,
                    out_shape=(dst_height, dst_width),
                    transform=dst_transform,
                    fill=fill,
                    default_value=default_value,
                    all_touched=all_touched,
                    dtype=dtype,
                )

Problem

• ml.fit_predict attempted to recast model predictions to unit16.
• this throws type error in rasterizer when prediction can't be recast

Solution

• If 'col' is included, use rasterio.dtypes.get_minimum_dtype to update dtype.
• The proposed solution is tesed and works for continuous and int dtypes

if col:
            shapes = (
                (geom, value) for geom, value in zip(dataframe.geometry, dataframe[col])
            )
            dtype=get_minimum_dtype(dataframe[col])

 else:
            shapes = dataframe.geometry.values

 varray = rasterize(
                shapes,
                out_shape=(dst_height, dst_width),
                transform=dst_transform,
                fill=fill,
                default_value=default_value,
                all_touched=all_touched,
                dtype=dtype,
            )

[mmann1123]

Resolved