ValueError：所有输入数组必须具有相同的形状

Question

I am trying to execute a code but got this Value error. Source.pkl file contains 100 elements as shown in array.png attached with this.

As shown in figure at index 0 there are 16 numpy arrays all with shape(1,59) so, first element has shape (16,1,59). At index 1 there are 11 numpy arrays all with shape(1,59) so, now shape of second element is (11,1,59) and so on for other elements also.

How can I have all the arrays with the same shape?

import numpy as np
from keras.layers import Input, LSTM, Dense
from keras.models import Model

pickle_in = open("source.pkl","rb")
predictions = np.load(pickle_in)[0:100]
in_predictions = np.stack(predictions)
pickle_in.close()

Answer 1

Here's two choices:

Append zero arrays to every list, until all of the lists have same size. In your case, assuming 16 is the largest size among all of the lists, you have to append (1, 59) shaped arrays to every list until the size of it is 16.

But, it's a waste of space. Another choice is:

Concatenate your lists to a large one, and then convert the large one into numpy array. Use another list to record which array was originally in which list. Here is a sample code, but I have not tested it:

pickle_in = open("source.pkl","rb")
predictions = np.load(pickle_in)[0:100]
wrap_list = []
index_map = []
for i in range(len(predictions)):
    pred = predictions[i]
    wrap_list += pred
    index_map += [i for _ in range(len(pred))]
in_predictions = np.array(wrap_list)
pickle_in.close()

I hope this could help you.