finalized model import and run

Author: sethjuarez

Digitz/MainWindow.xaml

@@ -50,7 +50,7 @@
             </Grid.RowDefinitions>
 
             <TextBlock Name="numberLabel"
-                 FontSize="150"
+                 FontSize="100"
                  Grid.Row="0" />
 
             <Grid Grid.Row="1">

Digitz/MainWindow.xaml.cs

@@ -27,6 +27,27 @@ public MainWindow()
             InitializeComponent();
         }
 
+        private void clearButton_Click(object sender, RoutedEventArgs e)
+        {
+            inkCanvas.Strokes.Clear();
+            numberLabel.Text = "";
+        }
+
+        private string Stringify(float[] data)
+        {
+            StringBuilder sb = new StringBuilder();
+            for (int i = 0; i < data.Length; i++)
+            {
+                if (i == 0) sb.Append("{\r\n\t");
+                else if (i % 28 == 0)
+                    sb.Append("\r\n\t");
+                sb.Append($"{data[i],3:##0}, ");
+
+            }
+            sb.Append("\r\n}\r\n");
+            return sb.ToString();
+        }
+
         private TFTensor GetWrittenDigit(int size)
         {
             RenderTargetBitmap b = new RenderTargetBitmap(
@@ -51,36 +72,26 @@ private TFTensor GetWrittenDigit(int size)
             // sanity check
             Console.Write(Stringify(data));
 
-            return new TFTensor(data);
+            return TFTensor.FromBuffer(new TFShape(1, data.Length), data, 0, data.Length);
         }
 
         private void recognizeButton_Click(object sender, RoutedEventArgs e)
         {
             var tensor = GetWrittenDigit(28);
-            string modelFile = "saved_model.pb";
-            var model = File.ReadAllBytes(modelFile);
-            var graph = new TFGraph();
-            graph.Import(model, "");
-        }
 
-        private string Stringify(float[] data)
-        {
-            StringBuilder sb = new StringBuilder();
-            for(int i = 0; i < data.Length; i++)
+            using (var graph = new TFGraph())
             {
-                if (i == 0) sb.Append("{\r\n\t");
-                else if (i % 28 == 0)
-                    sb.Append("\r\n\t");
-                sb.Append($"{data[i],3:##0}, ");
-                
+                graph.Import(File.ReadAllBytes("frozen.pb"));
+                var session = new TFSession(graph);
+                var runner = session.GetRunner();
+                runner.AddInput(graph["InputData"][0], tensor);
+                runner.Fetch(graph["Model/linear"][0]);
+                var output = runner.Run();
+                TFTensor result = output[0];
+                float[] p = ((float[][])result.GetValue(true))[0];
+                int guess = Array.IndexOf(p, p.Max());
+                numberLabel.Text = guess.ToString();
             }
-            sb.Append("\r\n}\r\n");
-            return sb.ToString();
-        }
-
-        private void clearButton_Click(object sender, RoutedEventArgs e)
-        {
-            inkCanvas.Strokes.Clear();
         }
     }
 }

LearnDigitz/LearnDigitz.py

@@ -2,6 +2,7 @@
 import sys
 import os
 import tensorflow as tf
+from tensorflow.python.tools import freeze_graph as freeze
 from tensorflow.examples.tutorials.mnist import input_data
 from datetime import *
 
@@ -26,7 +27,7 @@ def main(_):
 
     # Parameters
     learning_rate = 0.01
-    training_epochs = 50
+    training_epochs = 10
     batch_size = 100
     display_epoch = 1
     unique = datetime.now().strftime('%m-%d_%H_%M')
@@ -40,15 +41,16 @@ def main(_):
     y = tf.placeholder(tf.float32, [None, 10], name='LabelData')
 
     # Set model weights
-    W = tf.Variable(tf.zeros([784, 10]), name='Weights')
-    b = tf.Variable(tf.zeros([10]), name='Bias')
+    W = tf.Variable(tf.zeros([784, 10]), name='weights')
+    b = tf.Variable(tf.zeros([10]), name='bias')
 
     # Construct model and encapsulating all ops into scopes, making
     # Tensorboard's Graph visualization more convenient
     with tf.name_scope('Model'):
         # Model
-        #pred = tf.nn.softmax(tf.matmul(x, W) + b) # Softmax
-        pred = tf.matmul(x, W) + b # Softmax
+        #pred = tf.nn.softmax(tf.matmul(x, W) + b, name="model") # Softmax
+        pred = tf.add(tf.matmul(x, W), b, name="linear")  # linear combination
+
     with tf.name_scope('Loss'):
         # Minimize error using cross entropy
         # cost = tf.reduce_mean(-tf.reduce_sum(y * tf.log(pred), reduction_indices=1))
@@ -112,16 +114,29 @@ def main(_):
         print("Accuracy:", acc.eval({x: mnist.test.images, y: mnist.test.labels}))
 
         # saving model
-        builder = tf.saved_model.builder.SavedModelBuilder(export_path)
-        builder.add_meta_graph_and_variables(
-            sess,
-            [tf.saved_model.tag_constants.SERVING],
-            signature_def_map = {
-                "model": tf.saved_model.signature_def_utils.predict_signature_def(
-                    inputs = { "x": x },
-                    outputs = { "prediction": pred })
-        })
-        builder.save()
+        checkpoint = os.path.join(export_path, "model.ckpt")
+        saver = tf.train.Saver()
+        # checkpoint - variables
+        saver.save(sess, checkpoint)
+        # graph
+        tf.train.write_graph(sess.graph_def, export_path, "model.pb", as_text=False)
+        # freeze
+        # python "Python\Lib\site-packages\tensorflow\python\tools\freeze_graph.py" --input_graph=.\Profile.pb --input_checkpoint=.\Profile.ckpt --output_node_names=Output/Predictions,Output/Loss --output_graph=frozen.pb 
+        g = os.path.join(export_path, "model.pb")
+        frozen = os.path.join(export_path, "frozen.pb")
+        
+        freeze.freeze_graph(
+            input_graph = g, 
+            input_saver = "", 
+            input_binary = True, 
+            input_checkpoint = checkpoint, 
+            output_node_names = "Model/linear",
+            restore_op_name = "",
+            filename_tensor_name = "",
+            output_graph = frozen,
+            clear_devices = True,
+            initializer_nodes = ""
+        )
         print("Model saved!")
     exit(0)