使用字典替换numpy数组元素

Question

I have this numpy array

message = [ 97 98 114 97]

and this dictionary

codes = {97: '1', 98: '01', 114: '000'}

and I am now iterating through the numpy array and converting those numbers to the ones corresponding in the dictionary like this:

[codes[i] for i in message]

But this is really slow and takes a lot of memory, since I am creating a new list. Is there better approach? Maybe one in which I will have still the same numpy array, but with the new numbers like this?

message = [1 01 000 1]

Answer 1

Here's a NumPythonic solution using np.searchsorted -

np.asarray(codes.values())[np.searchsorted(codes.keys(),message)]

Please note that the output would be a NumPy array as well. If you would like to have a list output, wrap it with .tolist() -

np.asarray(codes.values())[np.searchsorted(codes.keys(),message)].tolist()

I would think the only bottleneck to this approach would be conversion to NumPy array with np.asarray() , as usually np.searchsorted is pretty efficient.

Sample run -

In [36]: message = [ 97, 98, 114, 97]

In [37]: codes = {97: '1', 98: '01', 114: '000'}

In [38]: [codes[i] for i in message]
Out[38]: ['1', '01', '000', '1']

In [39]: np.asarray(codes.values())[np.searchsorted(codes.keys(),message)]
Out[39]: 
array(['1', '01', '000', '1'], 
      dtype='|S3')

Answer 2

you can very simply update in place:

for i in np.arange(len(message)):
   message[i] = codes.get(message[i], message[i])

I'm sure there's a numpy specific syntax to iterate. But note, you don't preserve the byte string since messages is of int type and your dict values appear to be byte arrays. So you would need to copy to a new array to preserve type.