fix

Author: formath

demo/deep_model/freeze_graph.sh

@@ -23,5 +23,5 @@
 python ../../python/freeze_graph.py \
     --checkpoint_dir='./checkpoint' \
     --graph_pb='./model/predict_graph.pb' \
-    --output_node_names='forward/logit/add' \
+    --output_node_names='pctr' \
     --output_dir='./model'

python/deep_model.py

@@ -58,8 +58,8 @@ def forward(self, sparse_id, sparse_val):
 
             #dim = net.get_shape().as_list()[1]
             dim = self.hidden_layer[-1]
-            self.weight = tf.Variable(tf.truncated_normal([dim, 1], stddev=0.1), name='weight_out')
-            self.bias = tf.Variable(tf.truncated_normal([1], stddev=0.1), name='bias_out')
+            self.weight = tf.Variable(tf.truncated_normal([dim, 2], stddev=0.1), name='weight_out')
+            self.bias = tf.Variable(tf.truncated_normal([2], stddev=0.1), name='bias_out')
             with tf.variable_scope("logit"):
                 logits = tf.matmul(net, self.weight) + self.bias
 

python/predict_model.py

@@ -28,8 +28,7 @@
 
 # define loss
 logits, all_parameter = model.forward(sparse_id, sparse_val)
-train_label = tf.to_int64(label)
-cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits, labels=train_label, name='cross_entropy')
+softmax = tf.nn.softmax(logits, name='pctr')
 
 # save graph
 with tf.Session() as sess:

python/train.py

@@ -60,8 +60,6 @@
     train_label = tf.to_int64(train_label)
     cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits, labels=train_label, name='cross_entropy')
     loss = tf.reduce_mean(cross_entropy, name='loss')
-    auc, _ = tf.metrics.auc(predictions=logits, labels=train_label)
-    cost = tf.reduce_mean(loss, name='cost')
 
     # define optimizer
     print("Optimization algorithm: {}".format(FLAGS.optimizer))
@@ -82,7 +80,7 @@
         exit(1)
 
     global_step = tf.Variable(0, name='global_step', trainable=False)
-    train_op = optimizer.minimize(cost, global_step=global_step)
+    train_op = optimizer.minimize(loss, global_step=global_step)
 
     # to eval
     tf.get_variable_scope().reuse_variables()
@@ -92,7 +90,6 @@
     valid_label = tf.to_int64(valid_label)
     valid_cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=valid_logits, labels=valid_label)
     valid_loss = tf.reduce_mean(valid_cross_entropy)
-    valid_auc, _ = tf.metrics.auc(predictions=valid_logits, labels=valid_label)
 
     # saver
     checkpoint_file = FLAGS.checkpoint_dir + "/model.checkpoint"
@@ -113,11 +110,11 @@
     threads = tf.train.start_queue_runners(coord=coord, sess=sess)
     try:
         while not coord.should_stop():
-            _, step, train_loss_val, train_auc_val = sess.run([train_op, global_step, loss, auc])
+            _, step, train_loss_val = sess.run([train_op, global_step, loss])
             if step % FLAGS.steps_to_validate == 0:
-                valid_loss_val, valid_auc_val = sess.run([valid_loss, valid_auc])
-                print("Step: {}, train loss: {}, train auc: {}, valid loss: {}, valid auc: {}".format(
-                            step, train_loss_val, train_auc_val, valid_loss_val, valid_auc_val))
+                valid_loss_val = sess.run([valid_loss])
+                print("Step: {}, train loss: {}, valid loss: {}".format(
+                            step, train_loss_val, valid_loss_val))
     except tf.errors.OutOfRangeError:
         print("training done")
     finally:

src/deep_model.cc

@@ -136,21 +136,21 @@ int main(int argc, char* argv[]) {
   std::vector<tensorflow::Tensor> outputs;
 
   // Run the session, evaluating our "predict/add" operation from the graph
-  status = session->Run(inputs, {"forward/logit/add"}, {}, &outputs);
+  status = session->Run(inputs, {"pctr"}, {}, &outputs);
   if (!status.ok()) {
     std::cerr << status.ToString() << std::endl;
     return 1;
   } else {
     std::cout << "Run session successfully" << std::endl;
   }
 
-  // Grab the first output (we only evaluated one graph node: "predict/add")
+  // Grab the first output (we only evaluated one graph node: "pctr")
   // and convert the node to a scalar representation.
-  auto output_softmax = outputs[0].scalar<float>();
+  auto pctr = outputs[0].tensor<float, 2>()(0, 1);
 
   // Print the results
   std::cout << outputs[0].DebugString() << std::endl;
-  std::cout << "output value: " << output_softmax() << std::endl;
+  std::cout << "output value: " << pctr << std::endl;
 
   // Free any resources used by the session
   session->Close();