使用numpy.reshape（）添加尺寸

Question

I have a situation I'm trying to resolve. I have climate model data in an array with 480 monthly entries at the latitude and longitude grid points of this model (the particulars are irrelevant). That is, I have a numpy array with dimensions (480, lat, lon).

I need to create yearly averages at each spatial grid point. The way I would like to do this is by using numpy.reshape() to make a 4D array with dimensions (480/12, 12, lat, lon) (where I group the timesteps into groups of 12 and restack them, as it were). Then I would average along the second axis ( np.mean(dat_new, axis = 1) ).

The whole thing would be

dat_new = np.reshape(dat, (dat.shape[0]/12, 12, dat.shape[1], dat.shape[2]))
dat_annual_mean = np.mean(dat_new, axis = 1)

My question is: does reshape work this way? Would it rearrange things in the correct order? If not (or even if so) is there another (potentially less clumsy way to do this?)

Thanks.

Answer 1

You should first permute the dimensions so that your data is reshaped in the correct order.

dat_new=np.reshape(np.transpose(dat,(1,2,0)),
                   (dat.shape[1],dat.shape[2],dat.shape[0]//12, 12))
dat_annual_mean = np.mean(dat_new, axis = 3)

ie it will give you an array with dimensions (lat, lon, 480/12), which you can transpose further if necessary.

Answer 2

Simplified so we can see the whole array, this is what I think you are trying to do:

In [608]: arr = np.zeros((3*4,2),int)
In [609]: arr[:,0]=np.arange(12)
In [610]: arr
Out[610]: 
array([[ 0,  0],
       [ 1,  0],
       [ 2,  0],
       [ 3,  0],
       [ 4,  0],
       [ 5,  0],
       [ 6,  0],
       [ 7,  0],
       [ 8,  0],
       [ 9,  0],
       [10,  0],
       [11,  0]])
In [611]: arr.reshape(3,4,2)
Out[611]: 
array([[[ 0,  0],
        [ 1,  0],
        [ 2,  0],
        [ 3,  0]],

       [[ 4,  0],
        [ 5,  0],
        [ 6,  0],
        [ 7,  0]],

       [[ 8,  0],
        [ 9,  0],
        [10,  0],
        [11,  0]]])
In [612]: _.mean(axis=1)
Out[612]: 
array([[ 1.5,  0. ],
       [ 5.5,  0. ],
       [ 9.5,  0. ]])