Fixed Deep RNN Multi-gpu implementation to use tf

Fixed Deep RNN Multi-gpu implementation to use tf.python.ops.nn.rnn_cell DeviceWrapper class instead of implementing custom class to achieve the same result.  It would appear that the custom class shown with the code does not function.

14_recurrent_neural_networks.ipynb

@@ -107,11 +107,13 @@
    ]
   },
   {
-   "cell_type": "markdown",
+   "cell_type": "code",
+   "execution_count": null,
    "metadata": {
     "deletable": true,
     "editable": true
    },
+   "outputs": [],
    "source": [
     "# Basic RNNs"
    ]
@@ -1661,6 +1663,29 @@
     "outputs, states = tf.nn.dynamic_rnn(multi_layer_cell, X, dtype=tf.float32)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Alternatively, you can use the Tensorflow class DeviceWrapper - note you can define more than one layer per gpu"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "devices = [\"/gpu:0\", \"/gpu:1\", \"/gpu:2\"]  \n",
+    "cells = []\n",
+    "for dev in devices:\n",
+    "        cell = DeviceWrapper(rnn_cell.BasicRNNCell(num_units=n_neurons), dev)\n",
+    "        cells.append(cell)\n",
+    "\n",
+    "self.multiple_lstm_cells = tf.contrib.rnn.MultiRNNCell(cells, state_is_tuple=True)\n",
+    "outputs, states = tf.nn.dynamic_rnn(multi_layer_cell, X, dtype=tf.float32)"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 76,
@@ -2716,7 +2741,7 @@
   "language_info": {
    "codemirror_mode": {
     "name": "ipython",
-    "version": 3
+    "version": 3.0
    },
    "file_extension": ".py",
    "mimetype": "text/x-python",
@@ -2730,7 +2755,7 @@
    "navigate_menu": true,
    "number_sections": true,
    "sideBar": true,
-   "threshold": 6,
+   "threshold": 6.0,
    "toc_cell": false,
    "toc_section_display": "block",
    "toc_window_display": false