Update to run on torch 0.4

Trying to address two issues:
- return `action` as an item to fix runtime error
 ```
 UserWarning: volatile was removed and now has no effect. Use `with torch.no_grad():` instead.
 ```
may related to: SeanNaren/deepspeech.pytorch#277

- change number of actions from global to a local one which I believe that was your intention

1.dqn.ipynb

@@ -172,11 +172,11 @@
     "        super(DQN, self).__init__()\n",
     "        \n",
     "        self.layers = nn.Sequential(\n",
-    "            nn.Linear(env.observation_space.shape[0], 128),\n",
+    "            nn.Linear(num_inputs, 128),\n",
     "            nn.ReLU(),\n",
     "            nn.Linear(128, 128),\n",
     "            nn.ReLU(),\n",
-    "            nn.Linear(128, env.action_space.n)\n",
+    "            nn.Linear(128, num_actions)\n",
     "        )\n",
     "        \n",
     "    def forward(self, x):\n",
@@ -186,7 +186,7 @@
     "        if random.random() > epsilon:\n",
     "            state   = Variable(torch.FloatTensor(state).unsqueeze(0), volatile=True)\n",
     "            q_value = self.forward(state)\n",
-    "            action  = q_value.max(1)[1].data[0]\n",
+    "            action  = q_value.max(1)[1].item()\n",
     "        else:\n",
     "            action = random.randrange(env.action_space.n)\n",
     "        return action"