def test_select_less_scores_than_k_min(self, bandit_mock):
"""If min length is lt k_min, super().compute_rewards is used.
In this case, we expect the socres to be returned as they are.
"""
# Set-up
selector = BestKReward(['RF', 'SVM'])
bandit_mock.return_value = 'SVM'
# Run
choice_scores = {
'DT': [0.7, 0.75, 0.73],
'RF': [0.8],
'SVM': [0.9, 0.95, 0.93],
}
best = selector.select(choice_scores)
# Assert
assert best == 'SVM'
choice_rewards = {
'RF': [0.8],
'SVM': [0.9, 0.95, 0.93],
}
bandit_mock.assert_called_once_with(choice_rewards)
def test_select_more_scores_than_k_min(self, bandit_mock):
"""If min length is gt k_min, self.compute_rewards is used.
In this case, we expect the lower scores to be nan.
"""
# Set-up
selector = BestKReward(['RF', 'SVM'])
bandit_mock.return_value = 'SVM'
# Run
choice_scores = {
'DT': [0.7, 0.75, 0.73],
'RF': [0.8, 0.85, 0.83],
'SVM': [0.9, 0.95, 0.93],
}
best = selector.select(choice_scores)
# Assert
assert best == 'SVM'
choice_rewards_expected = {
'RF': [np.nan, 0.85, 0.83],
'SVM': [np.nan, 0.95, 0.93],
}
# TODO
(choice_rewards, ), _ = bandit_mock.call_args
assert choice_rewards.keys() == choice_rewards_expected.keys()
for k in choice_rewards:
assert k in choice_rewards_expected
np.testing.assert_array_equal(choice_rewards[k],
choice_rewards_expected[k])
def test_select_less_scores_than_k_min(self, bandit_mock):
"""If min length is lt k_min, super().compute_rewards is used.
In this case, we expect the socres to be returned as they are.
"""
# Set-up
selector = BestKReward(['RF', 'SVM'])
bandit_mock.return_value = 'SVM'
# Run
choice_scores = {
'DT': [0.7, 0.75, 0.73],
'RF': [0.8],
'SVM': [0.9, 0.95, 0.93],
}
best = selector.select(choice_scores)
# Assert
assert best == 'SVM'
choice_rewards = {
'RF': [0.8],
'SVM': [0.9, 0.95, 0.93],
}
bandit_mock.assert_called_once_with(choice_rewards)
def test_select_more_scores_than_k_min(self, bandit_mock):
"""If min length is gt k_min, self.compute_rewards is used.
In this case, we expect the lower scores to be zeroed.
"""
# Set-up
selector = BestKReward(['RF', 'SVM'])
bandit_mock.return_value = 'SVM'
# Run
choice_scores = {
'DT': [0.7, 0.75, 0.73],
'RF': [0.8, 0.85, 0.83],
'SVM': [0.9, 0.95, 0.93],
}
best = selector.select(choice_scores)
# Assert
assert best == 'SVM'
choice_rewards = {
'RF': [0., 0.85, 0.83],
'SVM': [0., 0.95, 0.93],
}
bandit_mock.assert_called_once_with(choice_rewards)
def test_compute_rewards_duplicates(self):
k = 3
choices = ['RF', 'SVM']
scores = [0.7, 0.8, 0.7, 0.1, 0.8, 0.7]
selector = BestKReward(choices, k=k)
rewards = selector.compute_rewards(scores)
np.testing.assert_array_equal(
np.sort(rewards), np.sort([np.nan, np.nan, np.nan, 0.7, 0.8, 0.8]))
def test_compute_rewards_gt_k(self):
# Set-up
selector = BestKReward(['RF', 'SVM'], k=3)
# Run
scores = [0.83, 0.8, 0.86, 0.9]
rewards = selector.compute_rewards(scores)
# Assert
np.testing.assert_array_equal(rewards, [0.83, np.nan, 0.86, 0.9])
def test_compute_rewards_lt_k(self):
# Set-up
selector = BestKReward(['RF', 'SVM'], k=3)
# Run
scores = [0.8, 0.9]
rewards = selector.compute_rewards(scores)
# Assert
assert rewards == scores
def test_compute_rewards_gt_k(self):
# Set-up
selector = BestKReward(['RF', 'SVM'], k=3)
# Run
scores = [0.83, 0.8, 0.86, 0.9]
rewards = selector.compute_rewards(scores)
# Assert
assert rewards == [0.83, 0.0, 0.86, 0.9]
def test___init__(self):
# Run
selector = BestKReward(['RF', 'SVM'], k=3)
# Assert
assert selector.choices == ['RF', 'SVM']
assert selector.k == 3