在Cython中迭代对象列表/集合的最快方法？

Question

I have this code for a neural network library I am building

for connection in self.backwardConnections:
    self._z += connection.value()

where connection is a cdef class Connection and backwardConnections is python list of connections.

I have two question

1. What is the fastest iterate through a Python list? (I could have also done it with the typical for int i in range (len (..)) )
2. If I abandon the python list approach, what collection type (eg numpy array, c++ vectors, etc) could hold my Connection objects and improve performance?

Answer 1

The for ... in ...: construction is the fastest way to iterate over a list in Python. You will also not significantly improve performance by switching to another data type. These two things are already very fast.

My guess is that you are going about optimizing this code the wrong way. Instead of looking at the looping mechanism, which likely takes up an insignificant fraction of your running time, why not look at optimizing whatever the call to .value() does?

There's also the possibility that Python is just too slow for what you're trying to do. I find such cases to be quite rare, though.