add placeholder for simnet-tf-freeze model

tools/simnet/train/tf/tf_simnet.py

@@ -68,7 +68,7 @@ def train(conf_dict):
         datafeed = datafeeds.TFPairwisePaddingData(conf_dict)
         input_l, input_r, neg_input = datafeed.ops()
         pos_score = net.predict(input_l, input_r)
-        output_prob = tf.identity(pos_score, name="output_preb")
+        output_prob = tf.identity(pos_score, name="output_prob")
         neg_score = net.predict(input_l, neg_input)
         loss_layer = utility.import_object(
             conf_dict["loss_py"], conf_dict["loss_class"])(conf_dict)
@@ -105,13 +105,29 @@ def freeze(conf_dict):
     """
     model_path = conf_dict["save_path"]
     freeze_path = conf_dict["freeze_path"]
-    saver = tf.train.import_meta_graph(model_path + '.meta')
-    with tf.Session() as sess:
-        saver.restore(sess, model_path)
-        var_graph_def = tf.get_default_graph().as_graph_def()
-        const_graph_def = graph_util.convert_variables_to_constants(sess, var_graph_def, ["output_prob"])
-        with tf.gfile.GFile(freeze_path, "wb") as f:
-            f.write(const_graph_def.SerializeToString())
+    training_mode = conf_dict["training_mode"]
+
+    graph = tf.Graph()
+    with graph.as_default():
+        net = utility.import_object(
+                conf_dict["net_py"], conf_dict["net_class"])(conf_dict)
+        test_l = dict([(u, tf.placeholder(tf.int32, [None, v], name=u))
+            for (u, v) in dict(conf_dict["left_slots"]).iteritems()])
+        test_r = dict([(u, tf.placeholder(tf.int32, [None, v], name=u))
+            for (u, v) in dict(conf_dict["right_slots"]).iteritems()])
+        pred = net.predict(test_l, test_r)
+        if training_mode == "pointwise":
+            output_prob = tf.nn.softmax(pred, -1, name="output_prob")
+        elif training_mode == "pairwise":
+            output_prob = tf.identity(pred, name="output_prob")
+
+        restore_saver = tf.train.Saver()
+    with tf.Session(graph=graph) as sess:
+        sess.run(tf.global_variables_initializer())
+        restore_saver.restore(sess, model_path)
+        output_graph_def = tf.graph_util.\
+            convert_variables_to_constants(sess, sess.graph_def, ["output_prob"])
+        tf.train.write_graph(output_graph_def, '.', freeze_path, as_text=False)
 
 def convert(conf_dict):
     """