I tried to replicate the issue there, but couldn't: the empty arrays were filled with zeros. So I'm not sure my guess there was correct. But while I was doing this, I ran timeit on the .astype vs dtype arg approaches, and the latter was significantly faster.
Hypothetical fix to #270.
I tried to replicate the issue there, but couldn't: the empty arrays were filled with zeros. So I'm not sure my guess there was correct. But while I was doing this, I ran timeit on the
.astype
vsdtype
arg approaches, and the latter was significantly faster.