Worth adding a test?

The other option is we have a safe_first_element function in cbook, not sure which is better here though.

@dstansby Test added.

@tacaswell I think it's safer to convert to ndarray instead of safe_first_element() because the passed object X needs to provide more array-like functionality than just indexing the first element:

if np.all(X[0] == X):
    return (X[0] == coords).astype(float)
kde = mlab.GaussianKDE(X, bw_method)

While I think, the first two lines would work for pandas.Series when using safe_first_element(), I'm not sure about GaussianKDE, at least its docstring claims to need a ndarray.