class TestAction(TestCase):
"""Tests related to the action"""
def setUp(self):
self.bandit = Bandit()
self.action = Action(self.bandit)
def test_action_value_is_gaussian(self):
"""
Verify that our action will get a correct value from N(mu, sigma)
- mu: the mean, here 0
- sigma: the variance, here 1
"""
results = [Action(self.bandit).value for x in range(5000)]
law_1 = get_percent([x for x in results if x > 0], results)
self.assertAlmostEqual(law_1, 50, delta=2, msg="%s%% of results are "
"above mu instead of around 51%%" % law_1)
law_2 = get_percent([x for x in results if x > -1 and x < 1], results)
self.assertAlmostEqual(law_2, 68, delta=2, msg="%s%% of results are "
"above mu instead of around 68%%" % law_2)
def test_gauss_with_numpy(self):
"""We use numpy, let's add some verifications"""
mu, sigma = 0, 1
value = np.random.normal(mu, sigma, 1000)
diff_mean = abs(mu - np.mean(value))
self.assertLess(diff_mean, 0.12, "diff mean = %s > 0.1" % diff_mean)
diff_vari = abs(sigma - np.std(value, ddof=1))
self.assertLess(diff_vari, 0.12, "diff vari = %s > 0.1" % diff_vari)
def test_action_can_play(self):
"""Let's play and check that action got a reward"""
action = self.action # shortcut
[action.play() for x in range(1000)]
lim = [action.mu - 3 * action.sigma, action.mu + 3 * action.sigma]
rewards_in_range = [r for r in action.rewards if lim[0] <= r <= lim[1]]
percent = get_percent(rewards_in_range, action.rewards)
self.assertAlmostEqual(percent, 85.0, delta=15, msg="%s%% instead of "
"99,7%% for a Normal Distribution" % percent)
# TODO verify why it comes to 86% sometimes !
def test_action_mean_reward(self):
"""Test that an action keep an history of rewards and that mean_reward
become closer to value after multiples plays"""
[self.action.play() for i in range(1, 1000)]
vari = abs(self.action.sigma - np.std(self.action.rewards, ddof=1))
self.assertLess(vari, 0.12, "diff variance = %s > 0.1" % vari)
