Reduction dtype notes - gchanan/pytorch GitHub Wiki

Same Input and Output Dtype

Function	NumPy Equivalent	NumPy Int	NumPy UInt	NumPy Float	PyTorch Int	PyTorch UInt	PyTorch Float
kthvalue	numpy.partition	same	same	same
max	numpy.amax	same	same	same
min	numpy.amin	same	same	same
mode	scipy.stats.mode	same	same	same

Integer Upcast

Function	NumPy Equivalent	NumPy Int	NumPy UInt	NumPy Float	PyTorch Int	PyTorch UInt	PyTorch Float
prod	numpy.prod	int64	uint64	same	same	same
sum	numpy.sum	int64	uint64	same	same	same
cumprod	numpy.cumprod	int64	uint64	same	same	same
cumsum	numpy.cumsum	int64	uint64	same	same	same

Integer Upcast to Float

Function	NumPy Equivalent	NumPy Int	NumPy UInt	NumPy Float	PyTorch Int	PyTorch UInt	PyTorch Float
mean	numpy.mean	float64	float64	same	N/I	N/I
median	numpy.median	float64	float64	same	same	same
norm	numpy.linalg.norm	float64	float64	same	N/I	N/I
std	num.std	float64	float64	same	N/I	N/I
var	numpy.var	float64	float64	same	N/I	N/I
renorm	None	N/A	N/A	N/A	N/I	N/I

• N/I: Not Implemented
• N/A: Not Applicable
• Same: Same input and output type