add tracking to welford

hopsflow/util.py

@@ -674,20 +674,39 @@ def integrate_array(
 
 
 class WelfordAggregator:
-    __slots__ = ["n", "mean", "_m_2"]
+    __slots__ = ["n", "mean", "_m_2", "_tracker"]
 
-    def __init__(self, first_value: np.ndarray):
+    def __init__(self, first_value: np.ndarray, track=False):
         self.n = 1
         self.mean = first_value
         self._m_2 = np.zeros_like(first_value)
 
-    def update(self, new_value: np.ndarray):
+        self._tracker: Optional[SortedList] = None
+        if track:
+            self._tracker = SortedList()
+
+    def update(self, new_value: np.ndarray, i: Optional[int] = None):
+        if self._tracker is not None:
+            if i is None:
+                raise ValueError("Tracking is enabled but no index was supplied.")
+
+            if self.has_sample(i):
+                return
+
+            self._tracker.add(i)
+
         self.n += 1
         delta = new_value - self.mean
         self.mean += delta / self.n
         delta2 = new_value - self.mean
         self._m_2 += np.abs(delta) * np.abs(delta2)
 
+    def has_sample(self, i: int) -> bool:
+        if self._tracker is None:
+            return False  # don't know
+
+        return i in self._tracker
+
     @property
     def sample_variance(self) -> np.ndarray:
         if self.n == 1:
@@ -778,9 +797,7 @@ def load_online_cache(save: str):
 
 
 def ensemble_mean_online(
-    args: Any,
-    save: str,
-    function: Callable[..., np.ndarray],
+    args: Any, save: str, function: Callable[..., np.ndarray], i: Optional[int] = None
 ) -> Optional[EnsembleValue]:
     path = get_online_data_path(save)
 
@@ -796,13 +813,13 @@ def ensemble_mean_online(
         with path.open("rb") as agg_file:
             aggregate = pickle.load(agg_file)
             if result is not None:
-                aggregate.update(result)
+                aggregate.update(result, i)
 
     else:
         if result is None:
             raise RuntimeError("No cache and no result.")
 
-        aggregate = WelfordAggregator(result)
+        aggregate = WelfordAggregator(result, i is not None)
 
     with path.open("wb") as agg_file:
         pickle.dump(aggregate, agg_file)