class AdaptiveSamplerTest(unittest.TestCase):
def setUp(self):
random.seed(123456)
tree = nx.DiGraph()
tree.add_edges_from([
(0, 1), (1, 2), (1, 3), (2, 4), (2, 5),
(0, 6), (6, 7), (0, 8)
])
self.assign_g_attrs(tree)
self.tree = tree
for t, nodes in enumerate([(0, ), (1, 6, 8), (2, 3, 7), (4, 5)]):
for n in nodes:
tree.node[n]['datetime'] = t
self.s = AdaptiveSampler(self.tree, B=3,
timespan_secs=1,
node_score_func=lambda p, c: p**2 / c)
def assign_g_attrs(self, tree):
for s, t in tree.edges_iter():
tree[s][t]['c'] = 1
for n in tree.nodes_iter():
tree.node[n]['r'] = 1
def test_sampler_init(self):
assert_equal(
# {0: 4, 1: 3, 2: 3, 6: 2},
[0, 2, 1, 6],
self.s.roots_sorted_by_upperbound
)
assert_equal(
1.0,
self.s.explore_proba
)
assert_equal(4, self.s.n_nodes_to_cover)
def test_update(self):
result_tree = nx.DiGraph()
result_tree.add_edges_from(
[(0, 1), (0, 6), (1, 3)]
)
self.assign_g_attrs(result_tree)
self.s.update(0, result_tree)
assert_equal(
0.5,
self.s.explore_proba
)
assert_equal(
set([0, 1]),
self.s.covered_nodes
)
assert_equal(
{1: 2 ** 2},
self.s.node2score
)
# case: score of node 1 increases
result_tree.add_edge(1, 2)
self.assign_g_attrs(result_tree)
self.s.update(0, result_tree)
assert_equal(
{1: 3 ** 2 / 2},
self.s.node2score
)
def test_update_border_case(self):
self.s.update(0, self.tree)
assert_equal(
set([0, 1, 2, 6]),
self.s.covered_nodes
)
assert_equal(0,
self.s.explore_proba)
def test_explore_proba(self):
assert_equal(1, self.s.explore_proba)
result_tree = nx.DiGraph()
result_tree.add_edges_from(
[(0, 1), (0, 6), (1, 3)]
)
self.assign_g_attrs(result_tree)
self.s.update(0, result_tree)
assert_almost_equal(2 / 4., self.s.explore_proba)
def test_take_via_explore(self):
r, tree = self.s.take()
assert_equal('explore',
self.s.random_action())
assert_equal(0, r)
assert_equal(
sorted([(0, 1), (0, 6), (0, 8)]),
sorted(tree.edges())
)
# on and on
r, tree = self.s.take()
assert_equal(2, r)
r, tree = self.s.take()
assert_equal(1, r)
def test_take_via_exploit(self):
# round 1
self.s.update(0, self.tree)
assert_equal('exploit',
self.s.random_action())
r, tree = self.s.take()
assert_equal(1, r)
# round 2
self.s.update(r, tree)
assert_true(r not in self.s.node2score)
