def _generate_prob_profiles(num_items, num_slots):
"""Another implementation of `distribution` for test purposes.
This function is the original implementation from Karl. jblespiau@ find it
useful to add it here as: 1) an additional test of our function 2) a check
that the initial code is correct too.
Args:
num_items: The number of items to distribute.
num_slots: The number of slots.
Returns:
A numpy array of shape [num_distributions, num_slots].
"""
if num_slots == 1:
return np.array([num_items])
num_rows = utils.n_choose_k(num_items + num_slots - 1, num_items)
distributions = np.empty([num_rows, num_slots])
ind = 0
for i in range(0, num_items + 1):
n_tmp = num_items - i
k_tmp = num_slots - 1
distributions_tmp = _generate_prob_profiles(n_tmp, k_tmp)
distributions[ind:ind +
np.shape(distributions_tmp)[0], :] = np.column_stack(
(np.array(
(np.ones(np.shape(distributions_tmp)[0]) * i)),
distributions_tmp))
ind = ind + np.shape(distributions_tmp)[0]
return distributions
def test_construction(self, num_players, num_strategies):
logging.info("Testing payoff table construction.")
table = heuristic_payoff_table.PayoffTable(num_players, num_strategies)
num_rows = utils.n_choose_k(num_players + num_strategies - 1,
num_players)
distributions = np.array(
list(utils.distribute(num_players, num_strategies)))
payoffs = np.full([int(num_rows), num_strategies], np.nan)
np.testing.assert_array_equal(
np.concatenate([distributions, payoffs], axis=1), table())
def test_distribution(self, num_items, num_slots, normalize):
distribution = list(utils.distribute(num_items, num_slots, normalize))
# Correct length.
# See https://en.wikipedia.org/wiki/Stars_and_bars_%28combinatorics%29.
self.assertLen(distribution,
utils.n_choose_k(num_items + num_slots - 1, num_items))
# No duplicates.
self.assertLen(distribution, len(set(distribution)))
sum_distribution = num_items if not normalize else 1
for d in distribution:
self.assertTrue(sum_distribution, sum(d))
self.assertTrue((np.asarray(d) >= 0).all())
