#68 new alternate constructors for front-end delay experiments

darc/designs.py

@@ -53,13 +53,14 @@ class BayesianAdaptiveDesignGeneratorDARC(DesignGeneratorABC):
     '''
 
     def __init__(self, DA=[0.], DB=DEFAULT_DB, RA=list(), RB=[100.],
-                 RA_over_RB=list(), PA=[1.], PB=[1.],
+                 RA_over_RB=list(), inter_reward_interval=list(),
+                 PA=[1.], PB=[1.],
                  max_trials=20,
                  NO_REPEATS=False):
         super().__init__()
 
-        self._input_type_validation(RA, DA, PA, RB, DB, PB, RA_over_RB)
-        self._input_value_validation(PA, PB, DA, DB, RA_over_RB)
+        self._input_type_validation(RA, DA, PA, RB, DB, PB, RA_over_RB, inter_reward_interval)
+        self._input_value_validation(PA, PB, DA, DB, RA_over_RB, inter_reward_interval)
 
         self._DA = DA
         self._DB = DB
@@ -68,6 +69,7 @@ def __init__(self, DA=[0.], DB=DEFAULT_DB, RA=list(), RB=[100.],
         self._PA = PA
         self._PB = PB
         self._RA_over_RB = RA_over_RB
+        self._inter_reward_interval = inter_reward_interval
         self.max_trials = max_trials
         self.NO_REPEATS = NO_REPEATS
 
@@ -101,8 +103,7 @@ def get_next_design(self, model):
             f'get_next_design() took: {time.time()-start_time:1.3f} seconds')
         return chosen_design_named_tuple
 
-
-    def _input_type_validation(self, RA, DA, PA, RB, DB, PB, RA_over_RB):
+    def _input_type_validation(self, RA, DA, PA, RB, DB, PB, RA_over_RB, inter_reward_interval):
         # NOTE: possibly not very Pythonic
         assert isinstance(RA, list), "RA should be a list"
         assert isinstance(DA, list), "DA should be a list"
@@ -111,6 +112,8 @@ def _input_type_validation(self, RA, DA, PA, RB, DB, PB, RA_over_RB):
         assert isinstance(DB, list), "DB should be a list"
         assert isinstance(PB, list), "PB should be a list"
         assert isinstance(RA_over_RB, list), "RA_over_RB should be a list"
+        assert isinstance(inter_reward_interval,
+                          list), "inter_reward_interval should be a list"
 
         # we expect EITHER values in RA OR values in RA_over_RB
         # assert (not RA) ^ (not RA_over_RB), "Expecting EITHER RA OR RA_over_RB as an"
@@ -120,8 +123,7 @@ def _input_type_validation(self, RA, DA, PA, RB, DB, PB, RA_over_RB):
         if not RA_over_RB:
             assert not RA is False, "If not providing list for RA_over_RB, we expect a list for RA"
 
-
-    def _input_value_validation(self, PA, PB, DA, DB, RA_over_RB):
+    def _input_value_validation(self, PA, PB, DA, DB, RA_over_RB, inter_reward_interval):
         '''Confirm values of provided design space specs are valid'''
         if np.any((np.array(PA) < 0) | (np.array(PA) > 1)):
             raise ValueError('Expect all values of PA to be between 0-1')
@@ -135,6 +137,10 @@ def _input_value_validation(self, PA, PB, DA, DB, RA_over_RB):
         if np.any(np.array(DB) < 0):
             raise ValueError('Expecting all values of DB to be >= 0')
 
+        if np.any(np.array(inter_reward_interval) < 0):
+            raise ValueError(
+                'Expecting all values of inter_reward_interval to be >= 0')
+
         if np.any((np.array(RA_over_RB) < 0) | (np.array(RA_over_RB) > 1)):
             raise ValueError('Expect all values of RA_over_RB to be between 0-1')
 
@@ -175,6 +181,8 @@ def _generate_all_possible_designs(self, assume_discounting=True):
         logging.debug(f'provided DB = {self._DB}')
         logging.debug(f'provided PB = {self._PB}')
         logging.debug(f'provided RA_over_RB = {self._RA_over_RB}')
+        logging.debug(
+            f'provided inter_reward_interval = {self._inter_reward_interval}')
 
         if not self._RA_over_RB:
             '''assuming we are not doing magnitude effect, as this is
@@ -246,6 +254,16 @@ def delayed(cls, max_trials):
         return cls(max_trials=max_trials,
                    RA=list(100*np.linspace(0.05, 0.95, 91)))
 
+    @classmethod
+    def delayed_frontend_delay(cls, max_trials):
+        '''Defaults for a front-end delay experiment. These typically use a
+        fixed reward ratio.
+        - inter_reward_interval = RA+RB'''
+        return cls(max_trials=max_trials,
+                   RA=[50.], RB=[100.],
+                   DA=[0., 7, 30, 30*3, 30*6, 365, 365*5],
+                   inter_reward_interval=[1, 7, 14, 30, 30*3, 30*6, 365])
+
     @classmethod
     def risky(cls, max_trials):
         prob_list = [0.1, 0.25, 0.5, 0.75, 0.8, 0.9]