Merge pull request #204 from NileshPS/master

Ch 16. Using native python list instead of deque for storing experience, increases sampling performance considerably.

16_reinforcement_learning.ipynb

@@ -1558,14 +1558,35 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 63,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [],
    "source": [
     "from collections import deque\n",
     "\n",
+    "\n",
+    "class ReplayMemory:\n",
+    "    \n",
+    "    def __init__(self, size):\n",
+    "        self.size = size\n",
+    "        self.buf= [None] * self.size\n",
+    "        self.index, self.count = 0, 0\n",
+    "        \n",
+    "    def append(self, data):\n",
+    "        self.buf[self.index] = data\n",
+    "        self.count = min(self.count + 1, self.size)\n",
+    "        self.index = (self.index + 1) % self.size\n",
+    "    \n",
+    "    def __getitem__(self, idx):\n",
+    "        return self.buf[idx]\n",
+    "    \n",
+    "    def __len__(self):\n",
+    "        return self.count\n",
+    "\n",
+    "\n",
+    "    \n",
     "replay_memory_size = 500000\n",
-    "replay_memory = deque([], maxlen=replay_memory_size)\n",
+    "replay_memory = ReplayMemory(replay_memory_size)\n",
     "\n",
     "def sample_memories(batch_size):\n",
     "    indices = np.random.permutation(len(replay_memory))[:batch_size]\n",