Why does Keras to_categorical behaves differently on [1, -1] and [2, -2]?
y = [1, -1, -1]
y_ = np_utils.to_categorical(y)
array([[ 0., 1.],
[ 0., 1.],
[ 0., 1.]])
y = [2, -2, -2]
y_ = np_utils.to_categorical(y)
array([[ 0., 0., 1.],
[ 0., 1., 0.],
[ 0., 1., 0.]])