How can I average an array of arrays in python?

Question

I have a simulation that runs over many times. Each time an array is produced and I insert it into a larger array keeping track of all the data. for example

record = []
for i in range(2):
     r = random.random()
     array = numpy.arange(20)*r
     array.shape = (10,2)
     record.append(array)
record = numpy.array(record)

which produces:

[[[  0.           0.88765927]
  [  1.77531855   2.66297782]
  [  3.55063709   4.43829637]
  [  5.32595564   6.21361492]
  [  7.10127419   7.98893346]
  [  8.87659274   9.76425201]
  [ 10.65191128  11.53957056]
  [ 12.42722983  13.3148891 ]
  [ 14.20254838  15.09020765]
  [ 15.97786693  16.8655262 ]]

 [[  0.           0.31394919]
  [  0.62789839   0.94184758]
  [  1.25579677   1.56974596]
  [  1.88369516   2.19764435]
  [  2.51159354   2.82554274]
  [  3.13949193   3.45344112]
  [  3.76739031   4.08133951]
  [  4.3952887    4.70923789]
  [  5.02318709   5.33713628]
  [  5.65108547   5.96503466]]]

Since each array represents a simulation in my program. I would like to average the 2 different arrays contained within record.

basically I would like an array with the same dimensions as array but it would be an average of all the individual runs.

I could obviously just loop over the arrays but there is a lot of data in my actual simulations so I think it would be very costly on time

example out put (obviously it wouldn't be zero):

average = [[0.0, 0.0]
           [0.0, 0.0]
           [0.0, 0.0]
           [0.0, 0.0]
           [0.0, 0.0]
           [0.0, 0.0]
           [0.0, 0.0]
           [0.0, 0.0]
           [0.0, 0.0]
           [0.0, 0.0]]

Answer 1

Your record array from the example above is three dimensional, with shape:

>>> record.shape
(2, 10, 2)

The first dimension corresponds to the 2 iterations of your experiment. To average them, you need to tell np.average to do its thing along axis=0

>>> np.average(record, axis=0)
array([[ 0.        ,  0.45688836],
       [ 0.91377672,  1.37066507],
       [ 1.82755343,  2.28444179],
       [ 2.74133015,  3.19821851],
       [ 3.65510686,  4.11199522],
       [ 4.56888358,  5.02577194],
       [ 5.4826603 ,  5.93954865],
       [ 6.39643701,  6.85332537],
       [ 7.31021373,  7.76710209],
       [ 8.22399044,  8.6808788 ]])

If you know beforehand how many simulations you are going to run, you are better off skipping the list thing altogether and doing something like this:

simulations, sim_rows, sim_cols = 1000000, 10, 2
record = np.empty((simulations, sim_rows, sim_cols))
for j in xrange(simulations) :
    record[j] = np.random.rand(sim_rows, sim_cols)

>>> np.average(record, axis=0)
[[ 0.50021935  0.5000554 ]
 [ 0.50019659  0.50009123]
 [ 0.50008591  0.49973058]
 [ 0.49995812  0.49973941]
 [ 0.49998854  0.49989957]
 [ 0.5002542   0.50027464]
 [ 0.49993122  0.49989623]
 [ 0.50024623  0.49981818]
 [ 0.50005848  0.50016798]
 [ 0.49984452  0.49999112]]