将字符串字典转换为 numpy arrays 的字典

Question

I have a dictionary structured similar to below.

test_dict = {1: 'I run fast', 2: 'She runs', 3: 'How are you?'}

What I'm trying to do is convert all the strings to 4x4 numpy arrays where each word is in it's own row and each letter occupies one cell of the array, populated with blanks for lines that wouldn't fill the entire row a whole row of blanks for sentences that are less than 4 words long. I also need to be able to tie it back to the ID, so the result needs to be in some format that can allow for referencing each array by it's ID later on.

I don't know of any pre built functions that can handle something like this, but I would be happy to be wrong. For now I've been trying to write a loop to handle it. Below is obviously incomplete because I'm stuck at the point of creating an array in the structure I would like.

for k in test_dict.keys():
    sentence = test_dict.getvalues(k)
    sentence_ascii = [ord(c) for c in sentence]
    sentence_array = np.array(sentence_ascii)

Answer 1

Is this what you mean?

{
    key: np.array([list(word.ljust(4)) for word in val.split()])
    for key, val in test_dict.items()
}

output:

{1: array([['I', ' ', ' ', ' '],
           ['r', 'u', 'n', ' '],
           ['f', 'a', 's', 't']], dtype='<U1'),
 2: array([['S', 'h', 'e', ' '],
           ['r', 'u', 'n', 's']], dtype='<U1'),
 3: array([['H', 'o', 'w', ' '],
           ['a', 'r', 'e', ' '],
           ['y', 'o', 'u', '?']], dtype='<U1')}

Answer 2

This'll ensure you have blank rows for sentences less than four words long.

new_dict = {i+1:np.empty((4,4),dtype='str') for i in range(len(test_dict))}
for k,v in test_dict.items():
    new_dict[k][:len(v.split())] = np.array([list(s)+['']*(4-len(s)) for s in v.split()])
    new_dict[k] = new_dict[k].view(np.int32)

Answer 3

You can use this to call your arrays using field 'ID':

dt=[('ID', '<i4'), ('sentences', object)]
new_dict = np.empty(len(test_dict), dtype=dt)
for i, (k, v) in enumerate(test_dict.items()):
  new_dict[i] = (k, np.pad(np.array([list("{:<4}".format(w)) for w in v.split(' ')]).view(np.int32), [(0,4-len(v.split(' '))),(0,0)]))

example output:

print(new_dict[new_dict['ID']==2]['sentences'])

[array([[ 83, 104, 101,  32],
   [114, 117, 110, 115],
   [  0,   0,   0,   0],
   [  0,   0,   0,   0]], dtype=int32)]