removed uncessary read/write from FQE callback

examples/d3rlpy_training_api_continuous_stochastic.py

@@ -29,7 +29,7 @@
 
 # Import evaluators
 from offline_rl_ope.api.d3rlpy.Scorers import (
-    ISEstimatorScorer
+    ISEstimatorScorer, QueryScorer
     )
 from offline_rl_ope.components.Policy import Policy
 
@@ -161,8 +161,8 @@
         discount=gamma, cache=is_callback, is_type="per_decision", 
         norm_weights=True)})
 
-    # for scr in fqe_scorers:
-    #     scorers.update({scr: QueryScorer(cache=fqe_callback, query_key=scr)})
+    for scr in fqe_scorers:
+        scorers.update({scr: QueryScorer(cache=fqe_callback, query_key=scr)})
 
     epoch_callback = EpochCallbackHandler([is_callback, fqe_callback])
 

examples/d3rlpy_training_api_discrete.py

@@ -146,8 +146,8 @@
                 }
             )
 
-    # for scr in fqe_scorers:
-    #     scorers.update({scr: QueryScorer(cache=fqe_callback, query_key=scr)})
+    for scr in fqe_scorers:
+        scorers.update({scr: QueryScorer(cache=fqe_callback, query_key=scr)})
 
     epoch_callback = EpochCallbackHandler([is_callback, fqe_callback, dva_callback])
 

src/offline_rl_ope/api/d3rlpy/Callbacks/DM.py

@@ -74,7 +74,7 @@ def run(self, algo: QLearningAlgoProtocol, epoch:int, total_step:int):
         _msg = "Must provide n_steps_per_epoch for FQE training"
         assert "n_steps_per_epoch" in self.__model_fit_kwargs, _msg
 
-        fqe.fit(
+        res = fqe.fit(
             self.__dataset, 
             evaluators=self.__scorers, 
             **self.__model_fit_kwargs, 
@@ -83,15 +83,7 @@ def run(self, algo: QLearningAlgoProtocol, epoch:int, total_step:int):
             experiment_name=f"EXP_{str(self.__cur_exp)}"
             )
 
-        res:Dict = {}
-        for scr in self.__scorers:
-            __file_path = os.path.join(
-                self.__logs_loc, "EXP_{}".format(self.__cur_exp), 
-                "{}.csv".format(scr))
-            lines = np.genfromtxt(__file_path, delimiter=',')
-            if len(lines.shape) == 1:
-                lines = lines.reshape(-1,1)
-            res[scr] = lines[-1:,-1].item()
+        res = res[-1][1]
         self.__cur_exp += 1
         self.cache = res