The reason this is happening is because BEST like all models call CommonDataProcessor to check for missing values. In this case however the dtype of the input is object and the underlying implementation relies on numpy.isnan, which wont't work with strings and objects. We can extend this by allowing isnan to work with objects:
The reason this is happening is because
BESTlike all models callCommonDataProcessorto check for missing values. In this case however thedtypeof the input isobjectand the underlying implementation relies onnumpy.isnan, which wont't work with strings and objects. We can extend this by allowingisnanto work with objects:This works because in IEEE specs
NaN != NaNwhichnumpy.nanimplements