One Hot Encoding using numpy [duplicate]
Usually, when you want to get a one-hot encoding for classification in machine learning, you have an array of indices.
import numpy as npnb_classes = 6targets = np.array([[2, 3, 4, 0]]).reshape(-1)one_hot_targets = np.eye(nb_classes)[targets]
The one_hot_targets
is now
array([[[ 0., 0., 1., 0., 0., 0.], [ 0., 0., 0., 1., 0., 0.], [ 0., 0., 0., 0., 1., 0.], [ 1., 0., 0., 0., 0., 0.]]])
The .reshape(-1)
is there to make sure you have the right labels format (you might also have [[2], [3], [4], [0]]
). The -1
is a special value which means "put all remaining stuff in this dimension". As there is only one, it flattens the array.
Copy-Paste solution
def get_one_hot(targets, nb_classes): res = np.eye(nb_classes)[np.array(targets).reshape(-1)] return res.reshape(list(targets.shape)+[nb_classes])
Package
You can use mpu.ml.indices2one_hot. It's tested and simple to use:
import mpu.mlone_hot = mpu.ml.indices2one_hot([1, 3, 0], nb_classes=5)
Something like :
np.array([int(i == 5) for i in range(10)])
Should do the trick.But I suppose there exist other solutions using numpy.
edit : the reason why your formula does not work : np.put does not return anything, it just modifies the element given in first parameter. The good answer while using np.put()
is :
a = np.zeros(10)np.put(a,5,1)
The problem is that it can't be done in one line, as you need to define the array before passing it to np.put()
You could use List comprehension:
[0 if i !=5 else 1 for i in range(10)]
turns to
[0,0,0,0,0,1,0,0,0,0]