updated the sample following the latest sample

kerryjiang · kerryjiang · commit c49b3b3353a2 · 2019-11-19T22:50:26.000-08:00
diff --git a/home/samples/DigitsRecognitionNeuralNetwork.ipynb b/home/samples/DigitsRecognitionNeuralNetwork.ipynb
@@ -51,6 +51,11 @@
     "int batch_size = 100;\n",
     "float learning_rate = 0.001f;\n",
     "int h1 = 200; // number of nodes in the 1st hidden layer\n",
+    "\n",
+    "Tensor x, y;\n",
+    "Tensor loss, accuracy;\n",
+    "Operation optimizer;\n",
+    "\n",
     "int display_freq = 100;\n",
     "float accuracy_test = 0f;\n",
     "float loss_test = 1f;"
@@ -91,25 +96,30 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "var graph = new Graph().as_default();\n",
-    "\n",
-    "// Placeholders for inputs (x) and outputs(y)\n",
-    "var x = tf.placeholder(tf.float32, shape: (-1, img_size_flat), name: \"X\");\n",
-    "var y = tf.placeholder(tf.float32, shape: (-1, n_classes), name: \"Y\");\n",
-    "\n",
-    "// Create a fully-connected layer with h1 nodes as hidden layer\n",
-    "var fc1 = fc_layer(x, h1, \"FC1\", use_relu: true);\n",
-    "// Create a fully-connected layer with n_classes nodes as output layer\n",
-    "var output_logits = fc_layer(fc1, n_classes, \"OUT\", use_relu: false);\n",
-    "// Define the loss function, optimizer, and accuracy\n",
-    "var logits = tf.nn.softmax_cross_entropy_with_logits(labels: y, logits: output_logits);\n",
-    "var loss = tf.reduce_mean(logits, name: \"loss\");\n",
-    "var optimizer = tf.train.AdamOptimizer(learning_rate: learning_rate, name: \"Adam-op\").minimize(loss);\n",
-    "var correct_prediction = tf.equal(tf.argmax(output_logits, 1), tf.argmax(y, 1), name: \"correct_pred\");\n",
-    "var accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32), name: \"accuracy\");\n",
-    "\n",
-    "// Network predictions\n",
-    "var cls_prediction = tf.argmax(output_logits, axis: 1, name: \"predictions\");"
+    "public Graph BuildGraph()\n",
+    "{\n",
+    "    var graph = new Graph().as_default();\n",
+    "\n",
+    "    // Placeholders for inputs (x) and outputs(y)\n",
+    "    x = tf.placeholder(tf.float32, shape: (-1, img_size_flat), name: \"X\");\n",
+    "    y = tf.placeholder(tf.float32, shape: (-1, n_classes), name: \"Y\");\n",
+    "\n",
+    "    // Create a fully-connected layer with h1 nodes as hidden layer\n",
+    "    var fc1 = fc_layer(x, h1, \"FC1\", use_relu: true);\n",
+    "    // Create a fully-connected layer with n_classes nodes as output layer\n",
+    "    var output_logits = fc_layer(fc1, n_classes, \"OUT\", use_relu: false);\n",
+    "    // Define the loss function, optimizer, and accuracy\n",
+    "    var logits = tf.nn.softmax_cross_entropy_with_logits(labels: y, logits: output_logits);\n",
+    "    loss = tf.reduce_mean(logits, name: \"loss\");\n",
+    "    optimizer = tf.train.AdamOptimizer(learning_rate: learning_rate, name: \"Adam-op\").minimize(loss);\n",
+    "    var correct_prediction = tf.equal(tf.argmax(output_logits, 1), tf.argmax(y, 1), name: \"correct_pred\");\n",
+    "    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32), name: \"accuracy\");\n",
+    "\n",
+    "    // Network predictions\n",
+    "    var cls_prediction = tf.argmax(output_logits, axis: 1, name: \"predictions\");\n",
+    "\n",
+    "    return graph;\n",
+    "}"
    ]
   },
   {
@@ -118,7 +128,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "void Train(Session sess, Datasets<MnistDataSet> mnist)\n",
+    "public void Train(Session sess, Datasets<MnistDataSet> mnist)\n",
     "{\n",
     "    // Number of training iterations in each epoch\n",
     "    var num_tr_iter = mnist.Train.Labels.shape[0] / batch_size;\n",
@@ -129,9 +139,12 @@
     "    float loss_val = 100.0f;\n",
     "    float accuracy_val = 0f;\n",
     "\n",
+    "    var sw = new Stopwatch();\n",
+    "    sw.Start();\n",
+    "\n",
     "    foreach (var epoch in range(epochs))\n",
     "    {\n",
-    "        Console.WriteLine($\"Training epoch: {epoch + 1}\");\n",
+    "        print($\"Training epoch: {epoch + 1}\");\n",
     "        // Randomly shuffle the training data at the beginning of each epoch \n",
     "        var (x_train, y_train) = mnist.Randomize(mnist.Train.Data, mnist.Train.Labels);\n",
     "\n",
@@ -142,35 +155,24 @@
     "            var (x_batch, y_batch) = mnist.GetNextBatch(x_train, y_train, start, end);\n",
     "\n",
     "            // Run optimization op (backprop)\n",
-    "            sess.run(optimizer, new FeedItem(x, x_batch), new FeedItem(y, y_batch));\n",
+    "            sess.run(optimizer, (x, x_batch), (y, y_batch));\n",
     "\n",
     "            if (iteration % display_freq == 0)\n",
     "            {\n",
     "                // Calculate and display the batch loss and accuracy\n",
-    "                var result = sess.run(\n",
-    "                    new[] { loss, accuracy },\n",
-    "                    new FeedItem(x, x_batch),\n",
-    "                    new FeedItem(y, y_batch));\n",
-    "                loss_val = result[0];\n",
-    "                accuracy_val = result[1];\n",
-    "                Console.WriteLine($\"iter {iteration.ToString(\"000\")}: Loss={loss_val.ToString(\"0.0000\")}, Training Accuracy={accuracy_val.ToString(\"P\")}\");\n",
+    "                (loss_val, accuracy_val) = sess.run((loss, accuracy), (x, x_batch), (y, y_batch));\n",
+    "                print($\"iter {iteration.ToString(\"000\")}: Loss={loss_val.ToString(\"0.0000\")}, Training Accuracy={accuracy_val.ToString(\"P\")} {sw.ElapsedMilliseconds}ms\");\n",
+    "                sw.Restart();\n",
     "            }\n",
     "        }\n",
     "\n",
     "        // Run validation after every epoch\n",
-    "        var results1 = sess.run(\n",
-    "            new[] { loss, accuracy },\n",
-    "            new FeedItem(x, mnist.Validation.Data),\n",
-    "            new FeedItem(y, mnist.Validation.Labels)\n",
-    "            );\n",
-    "            \n",
-    "        loss_val = results1[0];\n",
-    "        accuracy_val = results1[1];\n",
-    "        Console.WriteLine(\"---------------------------------------------------------\");\n",
-    "        Console.WriteLine($\"Epoch: {epoch + 1}, validation loss: {loss_val.ToString(\"0.0000\")}, validation accuracy: {accuracy_val.ToString(\"P\")}\");\n",
-    "        Console.WriteLine(\"---------------------------------------------------------\");\n",
+    "        (loss_val, accuracy_val) = sess.run((loss, accuracy), (x, mnist.Validation.Data), (y, mnist.Validation.Labels));\n",
+    "        print(\"---------------------------------------------------------\");\n",
+    "        print($\"Epoch: {epoch + 1}, validation loss: {loss_val.ToString(\"0.0000\")}, validation accuracy: {accuracy_val.ToString(\"P\")}\");\n",
+    "        print(\"---------------------------------------------------------\");\n",
     "    }\n",
-    "}\n"
+    "}"
    ]
   },
   {
@@ -179,20 +181,13 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "void Test(Session sess, Datasets<MnistDataSet> mnist)\n",
+    "public void Test(Session sess, Datasets<MnistDataSet> mnist)\n",
     "{\n",
-    "    var result = sess.run(\n",
-    "        new[] { loss, accuracy },\n",
-    "        new FeedItem(x, mnist.Test.Data),\n",
-    "        new FeedItem(y, mnist.Test.Labels)\n",
-    "    );\n",
-    "    \n",
-    "    loss_test = result[0];\n",
-    "    accuracy_test = result[1];\n",
-    "    Console.WriteLine(\"---------------------------------------------------------\");\n",
-    "    Console.WriteLine($\"Test loss: {loss_test.ToString(\"0.0000\")}, test accuracy: {accuracy_test.ToString(\"P\")}\");\n",
-    "    Console.WriteLine(\"---------------------------------------------------------\");\n",
-    "}\n"
+    "    (loss_test, accuracy_test) = sess.run((loss, accuracy), (x, mnist.Test.Data), (y, mnist.Test.Labels));\n",
+    "    print(\"---------------------------------------------------------\");\n",
+    "    print($\"Test loss: {loss_test.ToString(\"0.0000\")}, test accuracy: {accuracy_test.ToString(\"P\")}\");\n",
+    "    print(\"---------------------------------------------------------\");\n",
+    "}"
    ]
   },
   {
@@ -204,6 +199,15 @@
     "var mnist = await MnistModelLoader.LoadAsync(\"mnist\", true);"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "BuildGraph();"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
