You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In #234 and #235, @ns-rse noticed that passing float32 images doesn't work in newer versions of skan (which pass the image unmodified rather than a float64 copy. That's because the numba jitclass declaration has an array of fixed dtype of float64[:]:
so passing float32 data to it causes an error. But! It would be good to be able to support float32 data, not least because it uses half the RAM, for limited loss in precision. So we should figure out a way to do that, maybe switching out two classes, NBGraph and NBGraph32, for example, depending on the input data.
The text was updated successfully, but these errors were encountered:
In #234 and #235, @ns-rse noticed that passing float32 images doesn't work in newer versions of skan (which pass the image unmodified rather than a float64 copy. That's because the numba jitclass declaration has an array of fixed dtype of
float64[:]
:skan/src/skan/csr.py
Line 162 in 3d5da88
so passing float32 data to it causes an error. But! It would be good to be able to support float32 data, not least because it uses half the RAM, for limited loss in precision. So we should figure out a way to do that, maybe switching out two classes, NBGraph and NBGraph32, for example, depending on the input data.
The text was updated successfully, but these errors were encountered: