why is nonzero(arr != 0) faster than nonzero(arr) in python

NumPy uses a different way to read the array when it's boolean. It can grab like 64 or 256 values at the same time using SIMD (basically CPU tricks to process things in bulk). With integers, it just checks each one by one, so it's slower.

arr != 0 makes a bool array, then goes through the fast path, while arr stays as an int array and goes one by one.

Also, arr.view(bool) is the fastest because it skips making a new array and just reads the same memory as if it was boolean already.

Numpy has some optimizations documentation for further reading.