Make TFGAN tutorial Python 3 compatible.

research/gan/tutorial.ipynb

@@ -66,7 +66,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -101,6 +101,7 @@
     "import numpy as np\n",
     "import time\n",
     "import functools\n",
+    "from six.moves import xrange  # pylint: disable=redefined-builtin\n",
     "\n",
     "import tensorflow as tf\n",
     "\n",
@@ -1066,14 +1067,20 @@
     }
    ],
    "source": [
+    "def _get_next(iterable):\n",
+    "    try:\n",
+    "        return iterable.next()  # Python 2.x.x\n",
+    "    except AttributeError:\n",
+    "        return iterable.__next__()  # Python 3.x.x\n",
+    "\n",
     "# Run inference.\n",
     "predict_input_fn = _get_predict_input_fn(36, NOISE_DIMS)\n",
     "prediction_iterable = gan_estimator.predict(\n",
     "    predict_input_fn, hooks=[tf.train.StopAtStepHook(last_step=1)])\n",
-    "predictions = [prediction_iterable.next() for _ in xrange(36)]\n",
+    "predictions = [_get_next(prediction_iterable) for _ in xrange(36)]\n",
     "\n",
     "try: # Close the predict session.\n",
-    "    prediction_iterable.next()\n",
+    "    _get_next(prediction_iterable)\n",
     "except StopIteration:\n",
     "    pass\n",
     "\n",
@@ -1889,7 +1896,7 @@
     "assert images_to_eval % cat_dim == 0\n",
     "\n",
     "unstructured_inputs = tf.random_normal([images_to_eval, noise_dims-cont_dim])\n",
-    "cat_noise = tf.constant(range(cat_dim) * (images_to_eval // cat_dim))\n",
+    "cat_noise = tf.constant(list(range(cat_dim)) * (images_to_eval // cat_dim))\n",
     "cont_noise = tf.random_uniform([images_to_eval, cont_dim], -1.0, 1.0)\n",
     "\n",
     "with tf.variable_scope(infogan_model.generator_scope, reuse=True):\n",