Computational of the local dimension interrupted by empty return values from PCA

Hi, I have found a bug while computing the local dimension. For my data, this line

np.cumsum(pca.explained_variance_ratio_) > local_variance_threshold)[0][0] 

in plot.py crashes. 

I have fixed the issue by placing it in a try block and assigning the value -1 when estimating the local dimension fails.  I am submitting a PR and see you there. 
