Impossible ValueError. Value not in list while using numpy.unique

Question

My code throws a 'Y[i] is not in list' ValueError, even though this list is composed of the unique values in Y. I printed the list, the types of the list and the type of Y[i] but found no solution. Also, the error occurs irregularly.

To provide some context: I am trying to write a simple piece of code that checks if my K-Means classifier classified correctly. Because the cluster means are unlabeled ints, I want my output to be a matrix of integers such that C[h][y] represents the amount of times that my model classifies X[i] as h, while the actual label was y. Because the given labels are not necessarily integers, I try to assign them integers by creating a list of possible labels (V) and use the index of this list rather than the label itself.

The code (including debug prints):

    def classify(func, D):
        X = D[0]
        Y = D[1]
        V = list(np.unique(Y))    # <- V contains all values of Y
        print(V)
        print(type(V[0]),type(V[1]),type(V[2]))
        C = [V]
        for i in range(len(Y)):
            h = func(X[i])
            while len(C) < h+1:
                C.append(np.zeros(len(V)))
            if not Y[i] in V:
                print(type(Y[i]))
            y = V.index(Y[i])     # <- V does not contain Y[i]?
            C[h][y] += 1
        return np.array(C)

The output:

    [1.0, 2.0, 3.0]
    <class 'numpy.float64'> <class 'numpy.float64'> <class 'numpy.float64'>
    <class 'numpy.float64'>
    Traceback (most recent call last):
      File "leren6.py", line 38, in <module>
        main()
      File "leren6.py", line 18, in main
        C = classify(model, Data)
      File "leren6.py", line 33, in classify
        y = V.index(Y[i])
    ValueError: 3.0 is not in list

If you can fix this, you're officially awesome.

Answer 1

There isn't much information given (example function arguments which reproduce the bug would be helpful next time), but I suspect that this line is responsible:

C = [V]

The issue is that C[0] becomes another name for V. Hence, whenever the line C[h][y] += 1 is executed when h = 0, one item in V gets clobbered. Hence, while V may have started as [np.float64(1.0), np.float64(2.0), np.float64(3.0)] , it may not stay that way as it gets eroded through the loop.

Answer 2

may be a rounding problem, you're working with non integer values. Try to replace 1.0, 2.0... whith 1, 2... and see what happens