Update the training schedule for better convergence. The current...

Update the training schedule for better convergence. The current hyperparameters tend to diverge in Colab.

samples/core/guide/autograph.ipynb

@@ -10,7 +10,8 @@
       "collapsed_sections": [
         "Jxv6goXm7oGF"
       ],
-      "toc_visible": true
+      "toc_visible": true,
+      "include_colab_link": true
     },
     "kernelspec": {
       "name": "python3",
@@ -18,6 +19,16 @@
     }
   },
   "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "view-in-github",
+        "colab_type": "text"
+      },
+      "source": [
+        "[View in Colaboratory](https://colab.research.google.com/github/mdanatg/models/blob/master/samples/core/guide/autograph.ipynb)"
+      ]
+    },
     {
       "metadata": {
         "id": "Jxv6goXm7oGF",
@@ -740,7 +751,7 @@
         "@autograph.convert(recursive=True)\n",
         "def train(train_ds, test_ds, hp):\n",
         "  m = mlp_model((28 * 28,))\n",
-        "  opt = tf.train.MomentumOptimizer(hp.learning_rate, 0.9)\n",
+        "  opt = tf.train.AdamOptimizer(hp.learning_rate)\n",
         "  \n",
         "  # We'd like to save our losses to a list. In order for AutoGraph\n",
         "  # to convert these lists into their graph equivalent,\n",
@@ -802,7 +813,7 @@
       "source": [
         "with tf.Graph().as_default() as g:\n",
         "  hp = tf.contrib.training.HParams(\n",
-        "      learning_rate=0.05,\n",
+        "      learning_rate=0.005,\n",
         "      max_steps=500,\n",
         "  )\n",
         "  train_ds = setup_mnist_data(True, 50)\n",
@@ -837,4 +848,4 @@
       "outputs": []
     }
   ]
-}
+}
\ No newline at end of file