IsolationForest benchmark doesn't accept legacy multi-label data representation

[hrjn]

Description

While running benchmarks/bench_isolation_forest.py the following error occurs:

Traceback (most recent call last):
  File "bench_isolation_forest_OLD.py", line 67, in <module>
    lb.fit(X[:, 1])
  File "(...)/anaconda3/lib/python3.5/site-packages/sklearn/preprocessing/label.py", line 296, in fit
    self.y_type_ = type_of_target(y)
  File "(...)/anaconda3/lib/python3.5/site-packages/sklearn/utils/multiclass.py", line 250, in type_of_target
    raise ValueError('You appear to be using a legacy multi-label data'
ValueError: You appear to be using a legacy multi-label data representation.
Sequence of sequences are no longer supported; use a binary array or sparse matrix instead.

This seems to be a Python 3 specific issue (not reproducible on Python 2.7)

Versions

Darwin-15.6.0-x86_64-i386-64bit
Python 3.5.2 |Anaconda 4.2.0 (x86_64)| (default, Jul 2 2016, 17:52:12)
[GCC 4.2.1 Compatible Apple LLVM 4.2 (clang-425.0.28)]
NumPy 1.12.0
SciPy 0.18.1
Scikit-Learn 0.18.1

[hrjn]

Fix proposal in #8638