def intersection(df, pairwise=False, **subset_args):
"""
Counts the size of intersections of subsets of predicted examples.
E.g. count the overlap between the top k of two different models
Args:
df: the result of to_dataframe(), Predict steps of length n_steps
pairwise: when False, returns the mutual intersection between
all subsets. Otherwise returns an n_steps x n_steps matrix
whose i,j entry is the number of examples in the
intersection between the i and j step subsets.
**subset_args: arguments to be passed to model.y_subset()
for each predict step
Returns: the intersection, either an integer, if pairwise is False,
or a DataFrame, otherwise.
"""
indexes = map(lambda row: set(model.y_subset(row[1].step.get_result()['y'], **subset_args).index), df.iterrows())
if not pairwise:
return len(util.intersect(indexes))
else:
r = pd.DataFrame(index=df.index, columns=xrange(len(df)))
for i in xrange(len(df)):
r.values[i][i] = len(indexes[i])
for j in xrange(i+1, len(df)):
r.values[i][j] = len(indexes[i] & indexes[j])
return r
def test_subset_k():
assert set(y_subset(y, k=2).index) == set([1, 3])
def test_subset_dropna():
assert set(y_subset(y, dropna=True).index) == set([0, 1, 3, 4])
def test_subset_query():
assert set(y_subset(y, query="attr").index) == set([1])
def apply_y(df, fn, **kwargs):
return apply(df, lambda s: fn(model.y_subset(s.get_result()['y'], **kwargs)))
def test_subset_k():
assert set(y_subset(y, k=2).index) == set([1, 3])
def test_subset_dropna():
assert set(y_subset(y, dropna=True).index) == set([0, 1, 3, 4])
def test_subset_query():
assert set(y_subset(y, query='attr').index) == set([1])