The rasterio.features.sieve function docs states that it works with integer and float32 types. However, the code and actual behavior of rasterio.features.sieve only accepts interger types:

Running the code belowe always returns ValueError: image dtype must be one of: rasterio.int16, rasterio.int32, rasterio.uint8, rasterio.uint16, despite documentation saying it would accept float32.

import numpy
from rasterio.features import sieve

m = numpy.eye(5, dtype=numpy.float32)
res = sieve(m, 2)