Repository navigation
concatenating with empty list changes dtype to float64 #10135
Description
When concatenating anything with an empty list, the dtype of the result is augmented to float64.
MWE:
import numpy
print(numpy.concatenate([[0]]).dtype) # int64
print(numpy.concatenate([[0], [1]]).dtype) # int64
print(numpy.concatenate([[0], []]).dtype) # float64!
print(numpy.concatenate([[], [0]]).dtype) # float64!
Activity
pv commentedon Nov 30, 2017
nschloe commentedon Jan 12, 2018
Not sure why I closed this one here. The explanation makes sense, indeed
The
numpy.arraydocumentation says about the optionaldtypeargument tonumpy.array:This doesn't seem to be the case here: A clearly smaller dtype than
float64is, e.g.,int8. Takingint8as default input type for empty arrays would also prevent the needless upcasting as described in the original post in most cases. The only time where this would fail isrkern commentedon Jan 12, 2018
Empty arrays are a special case that doesn't really fall under that rule (and is rare enough that that part of the docstring is probably just distracting rather than enlightening). The general default for an array where there are no values to inspect is
float64. Compare toones(),zeros(), etc.eric-wieser commentedon Jan 12, 2018
I think @nschloe is questioning whether that's the right default for empty arrays, which IMO is:
nschloe commentedon Jan 14, 2018
Indeed. I'll open one.