FAIL: Doctest: sklearn.neighbors.approximate.LSHForest

[cosmos2006]

while buidling the scikit.learn, I encountered following error. Please have a look at it.

Regards
cosmos

FAIL: Doctest: sklearn.neighbors.approximate.LSHForest
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/lib64/python2.7/doctest.py", line 2201, in runTest
    raise self.failureException(self.format_failure(new.getvalue()))
AssertionError: Failed doctest test for sklearn.neighbors.approximate.LSHForest
  File "/usr/local/scikit-learn-master/sklearn/neighbors/approximate.py", line 110, in LSHForest

----------------------------------------------------------------------
File "/usr/local/scikit-learn-master/sklearn/neighbors/approximate.py", line 198, in sklearn.neighbors.approximate.LSHForest
Failed example:
    distances                                        # doctest: +ELLIPSIS
Expected:
    array([[ 0.069...,  0.149...],
           [ 0.229...,  0.481...],
           [ 0.004...,  0.014...]])
Got:
    array([[ 0.0693931 ,  0.14960959],
           [ 0.22932536,  0.49225586],
           [ 0.00489444,  0.01481913]])
----------------------------------------------------------------------
File "/usr/local/scikit-learn-master/sklearn/neighbors/approximate.py", line 202, in sklearn.neighbors.approximate.LSHForest
Failed example:

[amueller]

I haven't seen this one before.
I think we should set the random_state in the docstring or ignore the output. I'm not super familiar with the LSHForest. Actually I wonder how this test ever passed without setting the random_state. Ping @ogrisel @jnothman?
The result can vary widely, depending on the random state, so setting the precision differently is not really an option.

[jnothman]

Oh! Apparently this example is reporting the exact nearest neighbors due to very generous parameters (n_candidates, n_estimators are both high with respect to the tiny data). We should be reducing the parameters and fixing random_state.

[amueller]

"mostly" providing the nearest neighbors. It's not deterministic on my box.

@lesteve you wanna fix this one up for the release?

[lesteve]

@lesteve you wanna fix this one up for the release?

I can try to have a look