Ejemplo n.º 1
0
    def test_should_corrupt_0(self):
        """
        Test that the suggestions will get corrupted when we want them to be.
        Level 0 noise
        """
        dx_pattern = [False, False, False, False, False, False, False, False, False, False,
                      False, False, False, False, False, False, False, False, False, False]
        ax_pattern = [False, False, False, False, False, False, False, False, False, False,
                      False, False, False, False, False, False, False, False, False, False]

        self.user.study_condition = User.StudyConditions.DXAX_100
        self.user.save()
        self.suggestions_provider = Suggestions(self.user)

        generated_dx = []
        generated_ax = []
        for action_idx in range(20):
            self.user.refresh_from_db()
            self.user.number_state_requests += 1
            self.user.save()

            generated_dx.append(self.suggestions_provider._should_corrupt(Suggestions.DX_CORRUPT_IDX_OFFSETS))
            generated_ax.append(self.suggestions_provider._should_corrupt(Suggestions.AX_CORRUPT_IDX_OFFSETS))

        self.assertListEqual(dx_pattern, generated_dx)
        self.assertListEqual(ax_pattern, generated_ax)
Ejemplo n.º 2
0
 def _reset_rng(self, idx):
     """
     If idx = 0, reset the RNG to default. Else user the users' value. This
     refreshes the RNG values in the provider to simulate how we expect
     the RNG values to evolve as user requests come in
     """
     self.user.refresh_from_db()
     if idx == 0:
         self.user.rng_state = Suggestions.DEFAULT_RNG_SEED
         self.user.save()
         self.suggestions_provider = Suggestions(self.user)
         self.rng = np.random.default_rng(Suggestions.DEFAULT_RNG_SEED)
     else:
         self.suggestions_provider = Suggestions(self.user)
         self.rng = np.random.default_rng(self.user.rng_state)
Ejemplo n.º 3
0
    def setUp(self):
        self.sm = StudyManagement.get_default()

        # Create a user and log them in
        self.user = User.objects.create_user('test_user', 'test_user')
        self.user.study_condition = User.StudyConditions.DXAX_100
        self.user.start_condition = User.StartConditions.AT_COUNTER_OCCLUDING_ABOVE_MUG
        self.user.save()

        # # Log the user in on the client
        self.client.login(username='******', password='******')

        self.suggestions_provider = Suggestions(self.user)

        # Create an rng to shadow the user
        self.rng = np.random.default_rng(Suggestions.DEFAULT_RNG_SEED)
Ejemplo n.º 4
0
    def test_all_noise_conditions_see_noise(self):
        """Test that all the noise conditions actually do see noise in the first
        3 actions"""
        action_number_check_threshold = 3

        # Make sure we're running the same condition that we will have for the
        # study. This is where we went wrong last time
        self.sm.max_dx_suggestions = 3
        self.sm.max_ax_suggestions = 3
        self.sm.pad_suggestions = True
        self.sm.save()

        # Iterate through the study conditions
        for study_condition in [
            User.StudyConditions.DX_90, User.StudyConditions.AX_90, User.StudyConditions.DXAX_90,
            User.StudyConditions.DX_80, User.StudyConditions.AX_90, User.StudyConditions.DXAX_80,
        ]:
            self.user.study_condition = study_condition
            self.user.save()

            # Then run the same rigamarole of going testing noise without
            # padding
            for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
                encountered_noise = False
                self.user.number_state_requests = -1
                self.user.save()

                # There has to be some noise before the end of the optimal
                # action sequence
                for idx, (action, expected_values) in enumerate(action_sequence[:-1]):
                    self.user.number_state_requests += 1
                    self.user.save()

                    self._reset_rng(idx)

                    state = State(expected_values['server_state_tuple'])

                    # If DX should be shown, then calculate the noise probabilty
                    if self.user.show_dx_suggestions:
                        expected_suggestions = expected_values['dx_suggestions']
                        suggestions, _, will_corrupt = self._test_dx_suggestions(state, action, expected_suggestions, pad=True)
                        encountered_noise = encountered_noise or (will_corrupt and idx < action_number_check_threshold)

                    # If we should show action suggestions, then check for noise
                    if self.user.show_ax_suggestions:
                        expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]
                        suggestions, _, will_corrupt = self._test_ax_suggestions(state, action, expected_suggestions, pad=True)
                        encountered_noise = encountered_noise or (will_corrupt and idx < action_number_check_threshold)

                    # Check the rng state
                    self.user.refresh_from_db()
                    self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)

                # Check that we have encountered noise in this condition
                self.assertTrue(encountered_noise, f"Did not encounter noise in {study_condition}, {start_state_str}")
Ejemplo n.º 5
0
    def test_multiple_suggestions_without_padding(self):
        """Test the return of multiple suggestions without padding to ensure a
        constant number of suggestions at all timesteps"""
        self.sm.max_dx_suggestions = 2
        self.sm.max_ax_suggestions = 3
        self.sm.save()
        self.user.refresh_from_db()

        # Refresh the values in the suggestions provider
        self.suggestions_provider = Suggestions(self.user)

        for start_state, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                state = State(expected_values['server_state_tuple'])

                expected_suggestions = expected_values['dx_suggestions']
                _, _, will_corrupt = self._test_dx_suggestions(state, action, expected_suggestions, pad=False)
                self.assertFalse(will_corrupt)

                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]
                _, _, will_corrupt = self._test_ax_suggestions(state, action, expected_suggestions, pad=False)
                self.assertFalse(will_corrupt)
Ejemplo n.º 6
0
    def test_multiple_with_noise_and_padding(self):
        """Test multiple suggestions with padding and noise"""
        self.sm.max_dx_suggestions = 3
        self.sm.max_ax_suggestions = 4
        self.sm.pad_suggestions = True
        self.sm.save()

        self.user.study_condition = User.StudyConditions.DXAX_70
        self.user.save()
        self.user.refresh_from_db()
        self.assertEqual(0.3, self.user.noise_level)

        for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                self.user.number_state_requests += 1
                self.user.save()

                self._reset_rng(idx)
                state = State(expected_values['server_state_tuple'])

                # First we do the DX
                expected_suggestions = expected_values['dx_suggestions']

                self._test_dx_suggestions(state, action, expected_suggestions, pad=True)

                # Check the rng state
                self.user.refresh_from_db()
                self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)

                # Then we do the AX
                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]

                self._test_ax_suggestions(state, action, expected_suggestions, pad=True)

                # Check the rng state
                self.user.refresh_from_db()
                self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)
Ejemplo n.º 7
0
    def test_multiple_suggestions_with_padding(self):
        """Test multiple suggestions with padding, but no noise"""
        self.sm.max_dx_suggestions = 3
        self.sm.max_ax_suggestions = 3
        self.sm.pad_suggestions = True
        self.sm.save()
        self.user.refresh_from_db()

        # Refresh the values in the suggestions provider
        self.suggestions_provider = Suggestions(self.user)

        for start_state, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                state = State(expected_values['server_state_tuple'])

                expected_suggestions = expected_values['dx_suggestions']
                suggestions, _, will_corrupt = self._test_dx_suggestions(state, action, expected_suggestions, pad=True)
                self.assertFalse(will_corrupt)
                self.assertEqual(self.sm.max_dx_suggestions, len(suggestions))

                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]
                suggestions, _, will_corrupt = self._test_ax_suggestions(state, action, expected_suggestions, pad=True)
                self.assertFalse(will_corrupt)
                self.assertEqual(self.sm.max_ax_suggestions, len(suggestions))
Ejemplo n.º 8
0
    def test_noisy_ax_suggestions_no_padding(self):
        """Test the addition of noise to the AX suggestions, without adding
        padding"""
        self.user.study_condition = User.StudyConditions.DXAX_70
        self.user.save()
        self.assertEqual(0.3, self.user.noise_level)

        # Refresh the values in the provider
        self.suggestions_provider = Suggestions(self.user)

        for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                self.user.number_state_requests += 1
                self.user.save()

                self._reset_rng(idx)
                state = State(expected_values['server_state_tuple'])
                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]

                self._test_ax_suggestions(state, action, expected_suggestions, pad=False)

                # Check the rng state
                self.user.refresh_from_db()
                self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)
Ejemplo n.º 9
0
class SuggestionsTestCase(TestCase):
    """
    Test the suggestions
    """

    def setUp(self):
        self.sm = StudyManagement.get_default()

        # Create a user and log them in
        self.user = User.objects.create_user('test_user', 'test_user')
        self.user.study_condition = User.StudyConditions.DXAX_100
        self.user.start_condition = User.StartConditions.AT_COUNTER_OCCLUDING_ABOVE_MUG
        self.user.save()

        # # Log the user in on the client
        self.client.login(username='******', password='******')

        self.suggestions_provider = Suggestions(self.user)

        # Create an rng to shadow the user
        self.rng = np.random.default_rng(Suggestions.DEFAULT_RNG_SEED)

    def _reset_rng(self, idx):
        """
        If idx = 0, reset the RNG to default. Else user the users' value. This
        refreshes the RNG values in the provider to simulate how we expect
        the RNG values to evolve as user requests come in
        """
        self.user.refresh_from_db()
        if idx == 0:
            self.user.rng_state = Suggestions.DEFAULT_RNG_SEED
            self.user.save()
            self.suggestions_provider = Suggestions(self.user)
            self.rng = np.random.default_rng(Suggestions.DEFAULT_RNG_SEED)
        else:
            self.suggestions_provider = Suggestions(self.user)
            self.rng = np.random.default_rng(self.user.rng_state)

    def test_ordered_diagnoses(self):
        """Test the ordered diagnosis method from the Suggestions"""
        for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                state = State(expected_values['server_state_tuple'])
                expected_suggestions = expected_values['dx_suggestions']

                # Get the suggestions & test them
                suggestions = self.suggestions_provider.ordered_diagnoses(state, action, accumulate=True)
                self.assertListEqual(expected_suggestions, suggestions)

                suggestions = self.suggestions_provider.ordered_diagnoses(state, action)
                self.assertListEqual([expected_suggestions[0]], suggestions)

    def test_optimal_action(self):
        """Test the optimal actions method from the Suggestions"""
        for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                state = State(expected_values['server_state_tuple'])
                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]

                # Get the suggestions
                suggestions = self.suggestions_provider.optimal_action(state, action)

                # Test the suggestions
                self.assertListEqual(expected_suggestions, suggestions)

    def test_should_corrupt_3(self):
        """
        Test that the suggestions will get corrupted when we want them to be.
        Level 3 noise
        """
        dx_pattern = [False, True, False, False, True, False, False, False, True, False,
                      False, True, False, False, True, False, False, False, True, False]
        ax_pattern = [False, True, False, False, True, False, False, False, True, False,
                      False, True, False, False, True, False, False, False, True, False]

        self.user.study_condition = User.StudyConditions.DXAX_70
        self.user.save()
        self.suggestions_provider = Suggestions(self.user)

        generated_dx = []
        generated_ax = []
        for action_idx in range(20):
            self.user.refresh_from_db()
            self.user.number_state_requests += 1
            self.user.save()

            generated_dx.append(self.suggestions_provider._should_corrupt(Suggestions.DX_CORRUPT_IDX_OFFSETS))
            generated_ax.append(self.suggestions_provider._should_corrupt(Suggestions.AX_CORRUPT_IDX_OFFSETS))

        self.assertListEqual(dx_pattern, generated_dx)
        self.assertListEqual(ax_pattern, generated_ax)

    def test_should_corrupt_2(self):
        """
        Test that the suggestions will get corrupted when we want them to be.
        Level 2 noise
        """
        dx_pattern = [False, True, False, False, True, False, False, False, False, False,
                      False, True, False, False, True, False, False, False, False, False]
        ax_pattern = [False, True, False, False, True, False, False, False, False, False,
                      False, True, False, False, True, False, False, False, False, False]

        self.user.study_condition = User.StudyConditions.DXAX_80
        self.user.save()
        self.suggestions_provider = Suggestions(self.user)

        generated_dx = []
        generated_ax = []
        for action_idx in range(20):
            self.user.refresh_from_db()
            self.user.number_state_requests += 1
            self.user.save()

            generated_dx.append(self.suggestions_provider._should_corrupt(Suggestions.DX_CORRUPT_IDX_OFFSETS))
            generated_ax.append(self.suggestions_provider._should_corrupt(Suggestions.AX_CORRUPT_IDX_OFFSETS))

        self.assertListEqual(dx_pattern, generated_dx)
        self.assertListEqual(ax_pattern, generated_ax)

    def test_should_corrupt_1(self):
        """
        Test that the suggestions will get corrupted when we want them to be.
        Level 1 noise
        """
        dx_pattern = [False, True, False, False, False, False, False, False, False, False,
                      False, True, False, False, False, False, False, False, False, False]
        ax_pattern = [False, True, False, False, False, False, False, False, False, False,
                      False, True, False, False, False, False, False, False, False, False]

        self.user.study_condition = User.StudyConditions.DXAX_90
        self.user.save()
        self.suggestions_provider = Suggestions(self.user)

        generated_dx = []
        generated_ax = []
        for action_idx in range(20):
            self.user.refresh_from_db()
            self.user.number_state_requests += 1
            self.user.save()

            generated_dx.append(self.suggestions_provider._should_corrupt(Suggestions.DX_CORRUPT_IDX_OFFSETS))
            generated_ax.append(self.suggestions_provider._should_corrupt(Suggestions.AX_CORRUPT_IDX_OFFSETS))

        self.assertListEqual(dx_pattern, generated_dx)
        self.assertListEqual(ax_pattern, generated_ax)

    def test_should_corrupt_0(self):
        """
        Test that the suggestions will get corrupted when we want them to be.
        Level 0 noise
        """
        dx_pattern = [False, False, False, False, False, False, False, False, False, False,
                      False, False, False, False, False, False, False, False, False, False]
        ax_pattern = [False, False, False, False, False, False, False, False, False, False,
                      False, False, False, False, False, False, False, False, False, False]

        self.user.study_condition = User.StudyConditions.DXAX_100
        self.user.save()
        self.suggestions_provider = Suggestions(self.user)

        generated_dx = []
        generated_ax = []
        for action_idx in range(20):
            self.user.refresh_from_db()
            self.user.number_state_requests += 1
            self.user.save()

            generated_dx.append(self.suggestions_provider._should_corrupt(Suggestions.DX_CORRUPT_IDX_OFFSETS))
            generated_ax.append(self.suggestions_provider._should_corrupt(Suggestions.AX_CORRUPT_IDX_OFFSETS))

        self.assertListEqual(dx_pattern, generated_dx)
        self.assertListEqual(ax_pattern, generated_ax)

    def _test_dx_suggestions(self, state, action, expected_suggestions, pad):
        """Given state, action, and expected suggestions, test the suggestions"""
        alternatives = [x for x in constants.DIAGNOSES.keys() if x not in expected_suggestions]
        expected_suggestions = expected_suggestions[:self.sm.max_dx_suggestions]

        # Predict if we will see noise & then add noise accordingly
        will_corrupt = self.suggestions_provider._should_corrupt(Suggestions.DX_CORRUPT_IDX_OFFSETS)
        if will_corrupt:
            # Move the rng forward
            padded_suggestions = self.rng.choice(alternatives, size=len(expected_suggestions), replace=False).tolist()
        else:
            padded_suggestions = [x for x in expected_suggestions]

        # Then update if we need to pad
        if pad:
            alternatives = set(alternatives) - set(padded_suggestions)
            for _ in range(len(expected_suggestions), self.sm.max_dx_suggestions):
                padded_suggestions.append(self.rng.choice(list(sorted(alternatives))))
                alternatives.discard(padded_suggestions[-1])

        # Then get the suggestions & check if they match. Add additional checks
        # if they are corrupted
        suggestions = self.suggestions_provider.suggest_dx(state, action)
        self.assertGreaterEqual(self.sm.max_dx_suggestions, len(suggestions))

        # The first test
        if pad:
            self.assertEqual(self.sm.max_dx_suggestions, len(suggestions))
        else:
            self.assertEqual(len(expected_suggestions), len(suggestions))

        # The second test
        for idx, suggestion in enumerate(expected_suggestions):
            if will_corrupt:
                self.assertNotIn(suggestion, suggestions)
            else:
                self.assertIn(suggestion, suggestions)

        # The final test... Do we predict the padded values well?
        self.assertListEqual(padded_suggestions, suggestions)

        # Return the suggestions, the padded suggestions
        return suggestions, padded_suggestions, will_corrupt

    def _test_ax_suggestions(self, state, action, expected_suggestions, pad):
        """Given state, action, and expected suggestions, test the suggestions"""
        valid_actions_check = state.get_valid_actions()
        alternatives = [k for k, v in valid_actions_check.items() if (k not in expected_suggestions and v)]
        expected_suggestions = expected_suggestions[:self.sm.max_ax_suggestions]

        # Predict if we will see noise & then add noise accordingly
        will_corrupt = self.suggestions_provider._should_corrupt(Suggestions.AX_CORRUPT_IDX_OFFSETS)
        if will_corrupt:
            # Move the rng forward
            padded_suggestions = self.rng.choice(alternatives, size=len(expected_suggestions), replace=False).tolist()
        else:
            padded_suggestions = [x for x in expected_suggestions]

        # Then update if we need to pad
        if pad:
            alternatives = set(alternatives) - set(padded_suggestions)
            for _ in range(len(expected_suggestions), self.sm.max_ax_suggestions):
                padded_suggestions.append(self.rng.choice(list(sorted(alternatives))))
                alternatives.discard(padded_suggestions[-1])

        # Then get the suggestions & check if they match. Add additional checks
        # if they are corrupted
        suggestions = self.suggestions_provider.suggest_ax(state, action)
        self.assertGreaterEqual(self.sm.max_ax_suggestions, len(suggestions))

        # The first test
        if pad:
            self.assertEqual(self.sm.max_ax_suggestions, len(suggestions))
        else:
            self.assertEqual(len(expected_suggestions), len(suggestions))

        # The second test
        for idx, suggestion in enumerate(expected_suggestions):
            if will_corrupt:
                self.assertNotIn(suggestion, suggestions)
            else:
                self.assertIn(suggestion, suggestions)

        # The final test... Do we predict the padded values well?
        self.assertListEqual(padded_suggestions, suggestions)

        # Return the suggestions, the padded suggestions
        return suggestions, padded_suggestions, will_corrupt

    def test_multiple_suggestions_without_padding(self):
        """Test the return of multiple suggestions without padding to ensure a
        constant number of suggestions at all timesteps"""
        self.sm.max_dx_suggestions = 2
        self.sm.max_ax_suggestions = 3
        self.sm.save()
        self.user.refresh_from_db()

        # Refresh the values in the suggestions provider
        self.suggestions_provider = Suggestions(self.user)

        for start_state, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                state = State(expected_values['server_state_tuple'])

                expected_suggestions = expected_values['dx_suggestions']
                _, _, will_corrupt = self._test_dx_suggestions(state, action, expected_suggestions, pad=False)
                self.assertFalse(will_corrupt)

                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]
                _, _, will_corrupt = self._test_ax_suggestions(state, action, expected_suggestions, pad=False)
                self.assertFalse(will_corrupt)

    def test_noisy_dx_suggestions_no_padding(self):
        """Test the addition of noise to the DX suggestions, without adding
        padding"""
        self.user.study_condition = User.StudyConditions.DXAX_70
        self.user.save()
        self.assertEqual(0.3, self.user.noise_level)

        for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                self.user.number_state_requests += 1
                self.user.save()

                self._reset_rng(idx)
                state = State(expected_values['server_state_tuple'])
                expected_suggestions = expected_values['dx_suggestions']

                self._test_dx_suggestions(state, action, expected_suggestions, pad=False)

                # Check the rng state
                self.user.refresh_from_db()
                self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)

    def test_noisy_ax_suggestions_no_padding(self):
        """Test the addition of noise to the AX suggestions, without adding
        padding"""
        self.user.study_condition = User.StudyConditions.DXAX_70
        self.user.save()
        self.assertEqual(0.3, self.user.noise_level)

        # Refresh the values in the provider
        self.suggestions_provider = Suggestions(self.user)

        for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                self.user.number_state_requests += 1
                self.user.save()

                self._reset_rng(idx)
                state = State(expected_values['server_state_tuple'])
                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]

                self._test_ax_suggestions(state, action, expected_suggestions, pad=False)

                # Check the rng state
                self.user.refresh_from_db()
                self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)

    def test_multiple_suggestions_with_padding(self):
        """Test multiple suggestions with padding, but no noise"""
        self.sm.max_dx_suggestions = 3
        self.sm.max_ax_suggestions = 3
        self.sm.pad_suggestions = True
        self.sm.save()
        self.user.refresh_from_db()

        # Refresh the values in the suggestions provider
        self.suggestions_provider = Suggestions(self.user)

        for start_state, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                state = State(expected_values['server_state_tuple'])

                expected_suggestions = expected_values['dx_suggestions']
                suggestions, _, will_corrupt = self._test_dx_suggestions(state, action, expected_suggestions, pad=True)
                self.assertFalse(will_corrupt)
                self.assertEqual(self.sm.max_dx_suggestions, len(suggestions))

                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]
                suggestions, _, will_corrupt = self._test_ax_suggestions(state, action, expected_suggestions, pad=True)
                self.assertFalse(will_corrupt)
                self.assertEqual(self.sm.max_ax_suggestions, len(suggestions))

    def test_multiple_with_noise_and_padding(self):
        """Test multiple suggestions with padding and noise"""
        self.sm.max_dx_suggestions = 3
        self.sm.max_ax_suggestions = 4
        self.sm.pad_suggestions = True
        self.sm.save()

        self.user.study_condition = User.StudyConditions.DXAX_70
        self.user.save()
        self.user.refresh_from_db()
        self.assertEqual(0.3, self.user.noise_level)

        for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
            for idx, (action, expected_values) in enumerate(action_sequence):
                self.user.number_state_requests += 1
                self.user.save()

                self._reset_rng(idx)
                state = State(expected_values['server_state_tuple'])

                # First we do the DX
                expected_suggestions = expected_values['dx_suggestions']

                self._test_dx_suggestions(state, action, expected_suggestions, pad=True)

                # Check the rng state
                self.user.refresh_from_db()
                self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)

                # Then we do the AX
                expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]

                self._test_ax_suggestions(state, action, expected_suggestions, pad=True)

                # Check the rng state
                self.user.refresh_from_db()
                self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)

    def test_all_noise_conditions_see_noise(self):
        """Test that all the noise conditions actually do see noise in the first
        3 actions"""
        action_number_check_threshold = 3

        # Make sure we're running the same condition that we will have for the
        # study. This is where we went wrong last time
        self.sm.max_dx_suggestions = 3
        self.sm.max_ax_suggestions = 3
        self.sm.pad_suggestions = True
        self.sm.save()

        # Iterate through the study conditions
        for study_condition in [
            User.StudyConditions.DX_90, User.StudyConditions.AX_90, User.StudyConditions.DXAX_90,
            User.StudyConditions.DX_80, User.StudyConditions.AX_90, User.StudyConditions.DXAX_80,
        ]:
            self.user.study_condition = study_condition
            self.user.save()

            # Then run the same rigamarole of going testing noise without
            # padding
            for start_state_str, action_sequence in constants.OPTIMAL_ACTION_SEQUENCES.items():
                encountered_noise = False
                self.user.number_state_requests = -1
                self.user.save()

                # There has to be some noise before the end of the optimal
                # action sequence
                for idx, (action, expected_values) in enumerate(action_sequence[:-1]):
                    self.user.number_state_requests += 1
                    self.user.save()

                    self._reset_rng(idx)

                    state = State(expected_values['server_state_tuple'])

                    # If DX should be shown, then calculate the noise probabilty
                    if self.user.show_dx_suggestions:
                        expected_suggestions = expected_values['dx_suggestions']
                        suggestions, _, will_corrupt = self._test_dx_suggestions(state, action, expected_suggestions, pad=True)
                        encountered_noise = encountered_noise or (will_corrupt and idx < action_number_check_threshold)

                    # If we should show action suggestions, then check for noise
                    if self.user.show_ax_suggestions:
                        expected_suggestions = [] if idx == len(action_sequence)-1 else [action_sequence[idx+1][0]]
                        suggestions, _, will_corrupt = self._test_ax_suggestions(state, action, expected_suggestions, pad=True)
                        encountered_noise = encountered_noise or (will_corrupt and idx < action_number_check_threshold)

                    # Check the rng state
                    self.user.refresh_from_db()
                    self.assertEqual(Suggestions.get_next_rng_seed(self.rng), self.user.rng_state)

                # Check that we have encountered noise in this condition
                self.assertTrue(encountered_noise, f"Did not encounter noise in {study_condition}, {start_state_str}")
Ejemplo n.º 10
0
    def _simulate_user(self, user):
        """Simulate the user's experience"""
        actions = user.studyaction_set.order_by('start_timestamp')

        # Get the simulated user and reset them
        sim_user = User.objects.get(username='******')
        sim_user.study_condition = user.study_condition
        sim_user.start_condition = user.start_condition
        sim_user.rng_state = Suggestions.DEFAULT_RNG_SEED
        sim_user.number_state_requests = -1
        sim_user.save()

        # For each state visited by the user, simulate the suggestions
        prev_action = None
        start_state = State(user.start_condition.split('.'))
        schk = Suggestions()  # Just a means to get the optimal alternatives
        for action in actions:
            next_state = State(eval(action.start_state))

            # Get the next state and verify it
            if sim_user.study_condition in Command.V1_NOISE_USAGE_CONDITIONS:
                # Send an empty user, then update the json with the simulated
                # suggestions from the old way of doing things
                json = get_next_state_json(start_state.tuple, prev_action, None)
                json.update(self._v1_noise_get_suggestions_json(next_state, sim_user))
            else:
                json = get_next_state_json(start_state.tuple, prev_action, sim_user)

            assert tuple(json['server_state_tuple']) == next_state.tuple, \
                f"({start_state.tuple}, {prev_action}): {json['server_state_tuple']} != {next_state.tuple}"

            # Add the suggestions
            if user.show_dx_suggestions:
                action.dx_suggestions = json['dx_suggestions']

            if user.show_ax_suggestions:
                action.ax_suggestions = json['ax_suggestions']

            # Add a boolean if the data was corrupted
            if user.noise_level > 0:
                if (
                    user.show_dx_suggestions and
                    schk.ordered_diagnoses(start_state, prev_action)[0] not in action.dx_suggestions
                ):
                    action.corrupted_dx_suggestions = True

                if (
                    user.show_ax_suggestions and
                    schk.optimal_action(start_state, prev_action)[0] not in action.ax_suggestions
                ):
                    action.corrupted_ax_suggestions = True

            # Save the action
            action.save()
            prev_action = action.action
            start_state = State(eval(action.next_state))

        # Simulate the last suggestions call
        if sim_user.study_condition in Command.V1_NOISE_USAGE_CONDITIONS:
            json = get_next_state_json(start_state.tuple, prev_action, None)
            json.update(self._v1_noise_get_suggestions_json(start_state, sim_user))
        else:
            json = get_next_state_json(start_state.tuple, prev_action, sim_user)

        # Check the RNG state
        try:
            assert sim_user.rng_state == user.rng_state, \
                f"Mismatch end state... FML: {user}... {user.rng_state} != {sim_user.rng_state}"
        except Exception as e:
            self.stdout.write(self.style.ERROR(f"{e}"))
Ejemplo n.º 11
0
    def _v1_noise_get_suggestions_json(self, state, user):
        """Use the old style of garnering suggestions from the server"""
        suggestions_json = {}
        user.refresh_from_db()
        suggestions_provider = Suggestions(user)

        def add_noise_and_pad(suggestions, alternatives, pad):
            """The old definition of the noise + pad function"""
            pad = pad or len(suggestions)
            should_corrupt = (suggestions_provider.rng.uniform() < user.noise_level)
            if should_corrupt:
                suggestions = suggestions_provider.rng.choice(alternatives, size=len(suggestions), replace=False).tolist()

            alternatives = set(alternatives) - set(suggestions)
            while len(suggestions) < pad:
                suggestions.append(suggestions_provider.rng.choice(list(sorted(alternatives))))
                alternatives.discard(suggestions[-1])

            return suggestions

        # First add the DX suggestions
        if user.show_dx_suggestions:
            suggestions = suggestions_provider.ordered_diagnoses(state, None, accumulate=True)
        else:
            suggestions = []

        alternatives = [x for x in constants.DIAGNOSES.keys() if x not in suggestions]
        suggestions = add_noise_and_pad(
            suggestions[:user.study_management.max_dx_suggestions],
            alternatives,
            user.study_management.max_dx_suggestions if user.study_management.pad_suggestions else None
        )

        suggestions_json['dx_suggestions'] = suggestions

        # Update the RNG
        user.rng_state = Suggestions.get_next_rng_seed(suggestions_provider.rng)
        user.save()

        # Second the AX suggestions
        if user.show_ax_suggestions:
            suggestions = suggestions_provider.optimal_action(state, None)
        else:
            suggestions = []

        valid_actions = state.get_valid_actions()
        alternatives = [k for k, v in valid_actions.items() if (k not in suggestions and v)]
        suggestions  = add_noise_and_pad(
            suggestions[:user.study_management.max_ax_suggestions],
            alternatives,
            user.study_management.max_ax_suggestions if user.study_management.pad_suggestions else None
        )

        suggestions_json['ax_suggestions'] = suggestions

        # Update the RNG
        user.rng_state = Suggestions.get_next_rng_seed(suggestions_provider.rng)
        user.save()

        # Return the JSON
        return suggestions_json