从字符串中提取整数

Question

I want to create the target list for a clustering problem with numerous classes from the list of the class names (labels) across each instance of my dataset.

class_name = ['class_1','class_2','class_3','class_3','class_1','class_2',\
'class_2','class_1','class_1','class_2','class_1','class_3'] 

The target list should be like an array in the same length as the class_name list, where an integer is assigned to different class label. Which is the bets way to get this?

target = np.array([1, 2, 3, 3, 1, 2, 2, 1, 1, 2, 1, 3])

A class label (eg class_1) is in the form 'Xx_xx_xxx(A123)' or 'Xx_xx_xxx (A123)'. The text in the parenthesis is not fixed. The list type is 'unicode' .

Answer 1

You can use a list comprehension to split the strings on '_' characters, take the digit at index [1] , then convert to int

>>> target = np.array([int(i.split('_')[1]) for i in class_name])
>>> target
array([1, 2, 3, 3, 1, 2, 2, 1, 1, 2, 1, 3])

Answer 2

The first thing that you should do is get the classes in a standard format. From what you described above, if the classname is in the parens within the string, then you can use a regex to just get the classname.

import re
X = ['abc(class_1)', 'cde_(class_1)', 'def_(class_2)']
just_classes = [re.findall(r'\((.*)\)', thing)[0] for thing in X]
# ['class_1', 'class_1', 'class_2']

There are a few different approaches you can use here. If you're doing ml with the numpy, scipy stack, I'd suggest learning the sklearn library. It has a lot of useful machine learning and AI tools including encoding class names.

Using sklearn

from sklearn.preprocessing import LabelEncoder
class_names = ['class_1','class_2','class_3','class_3','class_1','class_2',\
        'class_2','class_1','class_1','class_2','class_1','class_3'] 

my_enc = LabelEncoder()
my_enc.fit(class_names)
encoded1 =  my_enc.transform(class_names)

No external library

classes = set(class_names)
d = {c:i for i,c in enumerate(classes)}
encoded2 = [d[c_name] for c_name in class_names]
print encoded1 #approach 1 gives numpy array
print encoded2 # approach 2 gives standard python list

Both of the approaches should work. It's not much code to implement on your own, but in general, I'd suggest looking at the sklearn preprocessing tools.