Use deque + Nearest neighbor

test.py

@@ -1,28 +1,33 @@
-import math
 import threading
 import time
+from collections import deque
 from threading import Thread
 
+import numpy as np
+
 
 class TemporalQueue:
     def __init__(self):
-        self.items = []
+        self.items = deque(maxlen=10)
-        self.timestamps = []
+        self.timestamps = deque(maxlen=10)
 
     def add(self, item, timestamp):
         self.items.append(item)
         self.timestamps.append(timestamp)
 
-    def get(self, timestamp=None):
+    def get_latest(self):
-        if timestamp is None:
+        return self.items[-1], self.timestamps[-1]
-            return self.items[-1], self.timestamps[-1]
 
-        # TODO(rcadene): implement nearest neighbor instead of hacky floor
+    def get(self, timestamp):
-        for idx, t in list(enumerate(self.timestamps))[::-1]:
+        timestamps = np.array(list(self.timestamps))
-            if math.floor(t) == math.floor(timestamp):
+        distances = np.abs(timestamps - timestamp)
-                return self.items[idx], t
+        nearest_idx = distances.argmin()
 
-        raise ValueError()
+        # print(float(distances[nearest_idx]))
+        if float(distances[nearest_idx]) > 1 / 5:
+            raise ValueError()
+
+        return self.items[nearest_idx], self.timestamps[nearest_idx]
 
     def __len__(self):
         return len(self.items)
@@ -46,7 +51,7 @@ class Policy:
     def inference_loop(self):
         prev_timestamp = None
         while not self.stop_event.is_set():
-            last_observation, last_timestamp = self.obs_queue.get()
+            last_observation, last_timestamp = self.obs_queue.get_latest()
 
             if prev_timestamp is not None and prev_timestamp == last_timestamp:
                 # in case inference ran faster than recording/adding a new observation in the queue