There seems to be a bug in the documentation of the classification of text documents: http://scikit-learn.org/stable/auto_examples/text/mlcomp_sparse_document_classification.html#sphx-glr-auto-examples-text-mlcomp-sparse-document-classification-py

The files are opened as utf-8 which leads to a bug. I have solved the issue changing "open(f)" into "open(f, encoding='latin1')".