Add low-quality experience filter operator #470

trinity/buffer/operators/__init__.py

@@ -9,6 +9,8 @@
         "reward_shaping_mapper": "trinity.buffer.operators.mappers.reward_shaping_mapper.RewardShapingMapper",
         "pass_rate_calculator": "trinity.buffer.operators.mappers.pass_rate_calculator.PassRateCalculator",
         "data_juicer": "trinity.buffer.operators.data_juicer_operator.DataJuicerOperator",
+        "low_quality_experience_filter": "trinity.buffer.operators.filters.low_quality_filter.LowQualityExperienceFilter",
+
     },
 )
 

trinity/buffer/operators/filters/low_quality_filter.py

@@ -0,0 +1,12 @@
+from typing import List, Tuple
+import math
+
+from trinity.buffer.operators.experience_operator import ExperienceOperator
+from trinity.common.experience import Experience
+
+
+class LowQualityExperienceFilter(ExperienceOperator):
+    def process(self, exps: List[Experience]) -> Tuple[List[Experience], dict]:
+        kept = [e for e in exps if e.reward is not None and e.reward == e.reward]
+
+        return kept, {"filtered_count": len(exps) - len(kept)}