如何使 numpy.reshape 用作特定的 for 循环

Question

I have a 4 dimension list that contains numpy arrays. The following general shape is expected for this list num_batches,50,batch_size,27 . The last batch (ie, num_batches = -1 ) usually has instances less than batch_size . So, if you loop over the first axis and print the dimension you will see something like: (50,100,27),(50,100,27),...,(50,100,27),(50,85,27) I would like to reshape it such that it has the dimensions 50,all_samples_num,27 , where all_samples_num is the sum of the second dimension of the printed dimensions. At first, I wrote the following for loop:

tt = np.empty((50,0,27))
for ttt in x:
    i = i + 1
    tmp = []
    for t in ttt:
        tmp.append(t)
    tt = np.append(tt, np.array(tmp),axis=1)

When I wanted to optimize the code, I used the following reshaping code:

xx = np.array(x[:-1].tolist()).reshape(50,-1,27)
xx = np.concatenate((xx,np.array(x[-1].tolist()).reshape(50,-1,27)),axis=0)

However, tt and xx are not equivalent. I tried C-order and F-order of reshape but still the resulting array is not equivalent to tt . How can I manipulate reshape such that it does give something identical to tt ?

Answer 1

"concatenate" is the API you are seeking. "reshape" changes the interpretation of the addressing without really changing the data.

import numpy as np

x1 = np.random.randint(100,size=(50,5,27))
x2 = np.random.randint(100,size=(50,15,27))
x3 = np.random.randint(100,size=(50,9,27))
x4 = np.random.randint(100,size=(50,3,27))

i = np.concatenate( (x1,x2,x3,x4), axis=1)
print(i.shape)