def test_basic_behaviour(self):
below_threshold = om.optimal_matching(['a'], ['b'], lambda x, y: 10,
lambda x: 15, 11)
self.assertEqual({
'lhs': {
'a': 'b'
},
'rhs': {
'b': 'a'
}
}, below_threshold)
above_threshold = om.optimal_matching(['a'], ['b'], lambda x, y: 10,
lambda x: 15, 9)
self.assertEqual({
'lhs': {
'a': None
},
'rhs': {
'b': None
}
}, above_threshold)
without_threshold = om.optimal_matching(['a'], ['b'], lambda x, y: 10,
lambda x: 15, None)
self.assertEqual({
'lhs': {
'a': 'b'
},
'rhs': {
'b': 'a'
}
}, below_threshold)
def test_basic_behaviour( self ):
below_threshold = om.optimal_matching(['a'], ['b'], lambda x,y: 10, lambda x: 15, 11)
self.assertEqual({'lhs': {'a': 'b'}, 'rhs': {'b': 'a'}}, below_threshold)
above_threshold = om.optimal_matching(['a'], ['b'], lambda x,y: 10, lambda x: 15, 9)
self.assertEqual({'lhs': {'a': None}, 'rhs': {'b': None}}, above_threshold)
without_threshold = om.optimal_matching(['a'], ['b'], lambda x,y: 10, lambda x: 15, None)
self.assertEqual({'lhs': {'a': 'b'}, 'rhs': {'b': 'a'}}, below_threshold)
def runTest( self ):
def cost_fct(x, y):
table = {'a': {}, 'b': {}}
table['a']['d'] = 3
table['a']['c'] = 1
table['b']['c'] = 2
table['b']['d'] = 5
return table[x][y]
assoc = om.optimal_matching(['a','b'], ['c', 'd'], cost_fct, lambda x: 10)
expected = {'lhs': {'a': 'd', 'b': 'c'}, 'rhs': {'d': 'a', 'c': 'b'}}
self.assertEqual(assoc, expected)
def runTest(self):
def cost_fct(x, y):
table = {'a': {}, 'b': {}}
table['a']['d'] = 3
table['a']['c'] = 1
table['b']['c'] = 2
table['b']['d'] = 5
return table[x][y]
assoc = om.optimal_matching(['a', 'b'], ['c', 'd'], cost_fct,
lambda x: 10)
expected = {'lhs': {'a': 'd', 'b': 'c'}, 'rhs': {'d': 'a', 'c': 'b'}}
self.assertEqual(assoc, expected)
def optimal_matching( lhs, rhs, nonmatch_threshold=30):
'''Associate two sets of tracklets using globally optimal nearest neighbor matching.
Returns a dictionary of associations of the form:
{'lhs': {lhs_instance: rhs_instance}, 'rhs': {rhs_instance: lhs_instance}}
In case of a nonmatch, the instance is associated with None.
lhs, rhs -- iterable of unique Tracklet instances
nonmatch_threshold -- distance threshold; below are potential matching partners
'''
# In case of euclidean distance based nearest neighbor matching, the nonmatch_cost is equivalent
# with a hard distance cutoff; therefore we set the cutoff to almost the nonmatch_cost. We shouldn't
# set it exactly to the cost, because there may be problems if some match cost has exactly the value of
# the nonmatch cost. We better give the optimization scheme some room to breathe.
match = _om.optimal_matching( lhs, rhs, Tracklet.distance, lambda x: nonmatch_threshold, nonmatch_threshold + 1)
return match
def optimal_matching( lhs, rhs, nonmatch_threshold=30):
'''Associate two sets of tracklets using globally optimal nearest neighbor matching.
Returns a dictionary of associations of the form:
{'lhs': {lhs_instance: rhs_instance}, 'rhs': {rhs_instance: lhs_instance}}
In case of a nonmatch, the instance is associated with None.
lhs, rhs -- iterable of unique Tracklet instances
nonmatch_threshold -- distance threshold; below are potential matching partners
'''
# In case of euclidean distance based nearest neighbor matching, the nonmatch_cost is equivalent
# with a hard distance cutoff; therefore we set the cutoff to almost the nonmatch_cost. We shouldn't
# set it exactly to the cost, because there may be problems if some match cost has exactly the value of
# the nonmatch cost. We better give the optimization scheme some room to breathe.
match = _om.optimal_matching( lhs, rhs, Tracklet.distance, lambda x: nonmatch_threshold, nonmatch_threshold + 1)
return match
def test_vertices_without_edges(self):
no_edges = om.optimal_matching(['a'], ['b'], lambda x, y: 20,
lambda x: 15, 10)
self.assertEqual({'lhs': {'a': None}, 'rhs': {'b': None}}, no_edges)
def test_both_empty(self):
assoc = om.optimal_matching([], [], lambda x, y: 0, lambda x: 0)
self.assertEqual(assoc, {'lhs': {}, 'rhs': {}})
def test_empty_rhs(self):
assoc = om.optimal_matching([1, 2], [], lambda x, y: 0, lambda x: 0)
expected = {'lhs': {1: None, 2: None}, 'rhs': {}}
self.assertEqual(assoc, expected)
def test_vertices_without_edges( self ):
no_edges = om.optimal_matching(['a'], ['b'], lambda x,y: 20, lambda x: 15, 10)
self.assertEqual({'lhs': {'a': None}, 'rhs': {'b': None}}, no_edges)
def test_both_empty(self):
assoc = om.optimal_matching([], [], lambda x,y: 0, lambda x: 0)
self.assertEqual(assoc, {'lhs': {}, 'rhs': {}})
def test_empty_rhs(self):
assoc = om.optimal_matching([1,2], [], lambda x,y: 0, lambda x: 0)
expected = {'lhs': {1: None, 2: None}, 'rhs': {}}
self.assertEqual(assoc, expected)