Remove create_serving_input_fn and export_strategies from lab to matc…

…h the solution

Fixes GoogleCloudPlatform#2441

courses/machine_learning/deepdive2/recommendation_systems/labs/walsmodel/model.py

@@ -132,45 +132,6 @@ def batch_predict(args):
             for best_items_for_user in topk.eval():
                 f.write(",".join(str(x) for x in best_items_for_user) + '\n')
 
-# Online prediction returns row and column factors as needed
-def create_serving_input_fn(args):
-    def for_user_embeddings(userId):
-        # All items for this user (for user_embeddings)
-        items = tf.range(args["nitems"], dtype = tf.int64)
-        users = userId * tf.ones(shape = [args["nitems"]], dtype = tf.int64)
-        ratings = 0.1 * tf.ones_like(tensor = users, dtype = tf.float32)
-        return items, users, ratings, tf.constant(value = True, dtype = tf.bool)
-
-    def for_item_embeddings(itemId):
-        # All users for this item (for item_embeddings)
-        users = tf.range(args["nusers"], dtype = tf.int64)
-        items = itemId * tf.ones(shape = [args["nusers"]], dtype = tf.int64)
-        ratings = 0.1 * tf.ones_like(tensor = users, dtype = tf.float32)
-        return items, users, ratings, tf.constant(value = False, dtype = tf.bool)
-
-    def serving_input_fn():
-        feature_ph = {
-            "userId": tf.placeholder(dtype = tf.int64, shape = 1),
-            "itemId": tf.placeholder(dtype = tf.int64, shape = 1)
-        }
-
-        (items, users, ratings, project_row) = \
-            tf.cond(pred = feature_ph["userId"][0] < tf.constant(value = 0, dtype = tf.int64),
-                true_fn = lambda: for_item_embeddings(feature_ph["itemId"]),
-                false_fn = lambda: for_user_embeddings(feature_ph["userId"]))
-        rows = tf.stack(values = [users, items], axis = 1)
-        cols = tf.stack(values = [items, users], axis = 1)
-        input_rows = tf.SparseTensor(indices = rows, values = ratings, dense_shape = (args["nusers"], args["nitems"]))
-        input_cols = tf.SparseTensor(indices = cols, values = ratings, dense_shape = (args["nusers"], args["nitems"]))
-
-        features = {
-            WALSMatrixFactorization.INPUT_ROWS: input_rows,
-            WALSMatrixFactorization.INPUT_COLS: input_cols,
-            WALSMatrixFactorization.PROJECT_ROW: project_row
-        }
-        return tf.contrib.learn.InputFnOps(features = features, labels = None, default_inputs = feature_ph)
-    return serving_input_fn
-
 def train_and_evaluate(args):
     train_steps = int(0.5 + (1.0 * args["num_epochs"] * args["nusers"]) / args["batch_size"])
     steps_in_epoch = int(0.5 + args["nusers"] / args["batch_size"])
@@ -186,8 +147,7 @@ def experiment_fn(output_dir):
             eval_input_fn = read_dataset(tf.estimator.ModeKeys.EVAL, args),
             train_steps = train_steps,
             eval_steps = 1,
-            min_eval_frequency = steps_in_epoch,
-            export_strategies = tf.contrib.learn.utils.saved_model_export_utils.make_export_strategy(serving_input_fn = create_serving_input_fn(args))
+            min_eval_frequency = steps_in_epoch
         )
 
     from tensorflow.contrib.learn.python.learn import learn_runner