Updated: used hybrid and corrected based on comments

Author: djarpin

introduction_to_applying_machine_learning/gluon_recommender_system/gluon_recommender_system.ipynb

@@ -119,7 +119,9 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "metadata": {},
+   "metadata": {
+    "collapsed": true
+   },
    "outputs": [],
    "source": [
     "!mkdir /tmp/recsys/\n",
@@ -379,7 +381,7 @@
    },
    "outputs": [],
    "source": [
-    "class MFBlock(gluon.Block):\n",
+    "class MFBlock(gluon.HybridBlock):\n",
     "    def __init__(self, max_users, max_items, num_emb, dropout_p=0.5):\n",
     "        super(MFBlock, self).__init__()\n",
     "        \n",
@@ -394,15 +396,15 @@
     "            self.dropout = gluon.nn.Dropout(dropout_p)\n",
     "            self.dense = gluon.nn.Dense(num_emb, activation='relu')\n",
     "            \n",
-    "    def forward(self, users, items):\n",
+    "    def hybrid_forward(self, F, users, items):\n",
     "        a = self.user_embeddings(users)\n",
     "        a = self.dense(a)\n",
     "        \n",
     "        b = self.item_embeddings(items)\n",
     "        b = self.dense(b)\n",
     "\n",
     "        predictions = self.dropout(a) * self.dropout(b)      \n",
-    "        predictions = nd.sum(predictions, axis=1)\n",
+    "        predictions = F.sum(predictions, axis=1)\n",
     "        return predictions"
    ]
   },
@@ -443,6 +445,7 @@
     "net.collect_params().initialize(mx.init.Xavier(magnitude=2.24),\n",
     "                                ctx=ctx,\n",
     "                                force_reinit=True)\n",
+    "net.hybridize()\n",
     "\n",
     "# Set optimization parameters\n",
     "opt = 'sgd'\n",
@@ -486,8 +489,7 @@
     "                with mx.autograd.record():\n",
     "                    output = net(user, item)               \n",
     "                    loss = loss_function(output, label)\n",
-    "                    loss.backward()\n",
-    "                net.collect_params().values()\n",
+    "                loss.backward()\n",
     "                trainer.step(batch_size)\n",
     "            except:\n",
     "                pass\n",
@@ -825,7 +827,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "We can see that our neural network and embedding model produces substantially better results (1.28 vs 1.65 on mean square error).\n",
+    "We can see that our neural network and embedding model produces substantially better results (~1.27 vs 1.65 on mean square error).\n",
     "\n",
     "For recommender systems, subjective accuracy also matters.  Let's get some recommendations for a random user to see if they make intuitive sense."
    ]
@@ -931,7 +933,9 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "metadata": {},
+   "metadata": {
+    "collapsed": true
+   },
    "outputs": [],
    "source": [
     "sagemaker.Session().delete_endpoint(predictor.endpoint)"

introduction_to_applying_machine_learning/gluon_recommender_system/recommender.py

@@ -50,6 +50,7 @@ def train(channel_input_dirs, hyperparameters, hosts, num_gpus, **kwargs):
     net.collect_params().initialize(mx.init.Xavier(magnitude=2.24),
                                     ctx=ctx,
                                     force_reinit=True)
+    net.hybridize()
 
     trainer = gluon.Trainer(net.collect_params(),
                             opt,
@@ -63,7 +64,7 @@ def train(channel_input_dirs, hyperparameters, hosts, num_gpus, **kwargs):
     return trained_net, customer_index, product_index
 
 
-class MFBlock(gluon.Block):
+class MFBlock(gluon.HybridBlock):
     def __init__(self, max_users, max_items, num_emb, dropout_p=0.5):
         super(MFBlock, self).__init__()
 
@@ -78,15 +79,15 @@ def __init__(self, max_users, max_items, num_emb, dropout_p=0.5):
             self.dropout = gluon.nn.Dropout(dropout_p)
             self.dense = gluon.nn.Dense(num_emb, activation='relu')
 
-    def forward(self, users, items):
+    def hybrid_forward(self, F, users, items):
         a = self.user_embeddings(users)
         a = self.dense(a)
 
         b = self.item_embeddings(items)
         b = self.dense(b)
 
         predictions = self.dropout(a) * self.dropout(b)      
-        predictions = nd.sum(predictions, axis=1)
+        predictions = F.sum(predictions, axis=1)
         return predictions
 
 
@@ -102,8 +103,7 @@ def execute(train_iter, test_iter, net, trainer, epochs, ctx):
                 with mx.autograd.record():
                     output = net(user, item)               
                     loss = loss_function(output, label)
-                    loss.backward()
-                net.collect_params().values()
+                loss.backward()
                 trainer.step(batch_size)
             except:
                 pass