Python Numpy TypeError: ufunc 'isfinite' not supported for the input types

Question

Here’s my code:

def topK(dataMat,sensitivity):
    meanVals = np.mean(dataMat, axis=0)
    meanRemoved = dataMat - meanVals
    covMat = np.cov(meanRemoved, rowvar=0)
    eigVals,eigVects = np.linalg.eig(np.mat(covMat))

I get the error in the title on the last line above. I suspect it has something to do with the datatype, so, here’s an image of the variable and datatype from the Variable Explorer in Spyder:

I’ve tried changing np.linalg.eig(np.mat(covMat)) to np.linalg.eig(np.array(np.mat(covMat))) and to np.linalg.eig(np.array(covMat)), nothing works. Any ideas?

Asked By: swabygw

||

Source

Answer 1

Your array has a dtype of object, but this should be some floating point dtype. Use e.g.

covMat = np.array(covMat, dtype=float)

to convert the dtype

Answered By: jmd_dk

Answer 2

If you got this error, that means the function you called used np.isfinite under the hood. Common functions that do so are np.allclose, np.isclose and linalg methods (as in the OP). As jmd_dk’s answer mentions, it usually means the dtype of your array is object.

A common fix is to cast the array to float dtype.

covMat = covMat.astype(float)

Another common case is when concatenation / stack goes wrong or when the array comes from another datatype (e.g. pandas dataframes). In that case, try converting to a list first and then cast to np.ndarray.

arr = np.array(arr.tolist())

For example, for the following case, straightforward recasting won’t work but converting to a list and then into an ndarray works.

arr = pd.Series([np.array([1,2]), np.array([3,4]), np.array([5,6])]).values
# array([array([1, 2]), array([3, 4]), array([5, 6])], dtype=object)

np.array(arr, dtype=float)                  # <---- error
np.array(arr.tolist(), dtype=float)         # <---- no error

Answered By: cottontail

Python Numpy TypeError: ufunc 'isfinite' not supported for the input types

Question:

Answers: