Question

I'm learning python, numpy and machine learning. I'm trying to set up neural network from scratch and I have a problem.

I have some outputs e.g [[2], [4], [1]] and I'm trying to create mask for it that will look like this

[
[0 0 1 0 0]
[0 0 0 0 1]
[0 1 0 0 0] 
]

for now I'm using following code:

tmpY = np.array(Y)
tmp = np.zeros([m, 10])
for i in range (0, m):
    index = tmpY[i][0]
    tmp[i][index] = 1

But I think there is a cleaner way.

Edit:

Thanks guys for your help. I think I've found solution that will work best for me

C = np.array([[0], [2], [4], [2], [4], [1] ,[3], [8], [5], [3], [1], [2]])
np.eye(C.shape[0], np.amax(C) + 1, dtype=int)[C.flatten()]

[[1 0 0 0 0 0 0 0 0]
 [0 0 1 0 0 0 0 0 0]
 [0 0 0 0 1 0 0 0 0]
 [0 0 1 0 0 0 0 0 0]
 [0 0 0 0 1 0 0 0 0]
 [0 1 0 0 0 0 0 0 0]
 [0 0 0 1 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 1]
 [0 0 0 0 0 1 0 0 0]
 [0 0 0 1 0 0 0 0 0]
 [0 1 0 0 0 0 0 0 0]
 [0 0 1 0 0 0 0 0 0]]

I'll leave it here in case someone else will look it.

Answer 1

您的解决方案是正确的，这只是一个清洁一点的版本

indices = [[2],[4],[1]]
mask = np.zeros((m,10),dtype=np.uint8)

for i,indices in enumerate(indices): mask[i,indices] = 1

不确定从何处获得indices数组，但是您有某种条件想要掩盖原始图像，可以这样做：

original = np.random.uniform((100,100))

mask = np.zeros(original.shape,dtype=np.uint8)
mask[condition(original)] = 1 # eg mask[original < 0.5] = 1

Answer 2

sklearn has a class that can help you do this. You can use OneHotEncoder to create the mask as per the documentation

In your example

from sklearn.preprocessing import OneHotEncoder
enc = OneHotEncoder(handle_unknown='ignore')
X = [[2], [4], [1]]
enc.fit(X)

Then the output looks like:

enc.transform(X).toarray()
array([[0., 1., 0.],
       [0., 0., 1.],
       [1., 0., 0.]])

EDIT: You'll notice the output here has 3 elements for each transformed entry; this is because category 3 does not appear in the data we use to fit OneHotEncoder

How to create mask with one at specified index?

2 个答案: