Numpy - How can I transform this array without using python loops?-CodePudding

My input is a list of y_true labels, where the element in position i contains a value in the range of 0..len(classes) and depicts what class that element of the data set truly is. i ranges from 0 to len(data). Example below:

# 5 elements in data, 3 classes, all of which had representation in the data:
y_true = [0,2,1,0,1]

I want my output to be a list of lists that islen(data) by len(classes), where inner list i would have a 1 in the position of y_true[i], and 0 in the other len(classes)-1 slots, example:

#same configuration as the previous example
y_true = [0,2,1,0,1]  
result = [[1,0,0],[0,0,2],[0,1,0],[1,0,0],[0,1,0]]

Here's how I'm initilazing result:

result = np.zeros((len(y_true), max(y_true) 1))

However I haven't been able to make any further progress with this issue. I tried using add.at(result, y_true, 1) and this with y_true's shape flipped, but neither produced the result I wanted. What fuction(s) can achieve what I'm trying to do here?

Edit: For better clarity on what I want to achieve, I made it using a for loop:

result = np.zeros((len(y_true), max(y_true) 1))
for x in range(4):
  result[x][y_true[x]] = 1

CodePudding user response：

You can use fancy indexing:

result = np.zeros((len(y_true), max(y_true) 1), dtype=int)
result[np.arange(len(y_true)), y_true] = 1

output:

array([[1, 0, 0],
       [0, 0, 1],
       [0, 1, 0],
       [1, 0, 0],
       [0, 1, 0]])

alternative

an interesting alternative might be to use pandas.get_dummies:

import pandas as pd
result = pd.get_dummies(y_true).to_numpy()

output:

array([[1, 0, 0],
       [0, 0, 1],
       [0, 1, 0],
       [1, 0, 0],
       [0, 1, 0]], dtype=uint8)