python - Creating DOK matrix in scipy

Question

I have numpy arrays I need to convert to DOK format. Why does this work:

x = np.asarray([1, 0, 0])
y = scipy.sparse.dok_matrix(scipy.sparse.csr_matrix(x))

But this doesn't:

x = np.asarray([1, 0, 0])
y = scipy.sparse.dok_matrix(x)

which throws the error: "TypeError: Expected rank <=2 dense array or matrix."

I don't find anything in Scipy's docs that suggest I shouldn't be able to convert an array to a dok matrix directly.

Answer 1

csr_matrix accepts a 1-d array because it (eventually) passes the array argument through numpy.atleast_2d. dok_matrix doesn't do that--it requires a 2-d array. (I don't know why the behavior is different.)

To work around this, convert your 1-d input to 2-d before passing it to dok_matrix. E.g.

y = scipy.sparse.dok_matrix(np.atleast_2d(x))