try numba

src/models/esmm/model.py

@@ -96,9 +96,7 @@ def _bytes_feature(value):
 
 
 @numba.jit(nopython=True, parallel=True)
-def model_predict2(device_id, diary_ids, device_dict, diary_dict, predict_fn):
-    time_1 = timeit.default_timer()
-    device_info, diary_lst = device_diary_fe(device_id, diary_ids, device_dict, diary_dict)
+def _make_example(device_info, diary_lst):
     int_columns = [
         "active_type", "active_days", "card_id", "is_pure_author", "is_have_reply", "is_have_pure_reply", "content_level",
         "topic_num", "favor_num", "vote_num"
@@ -121,11 +119,19 @@ def model_predict2(device_id, diary_ids, device_dict, diary_dict, predict_fn):
         for col in float_columns:
             features[col] = _float_feature(float(tmp[col]))
         for col in str_columns:
-            features[col] = _bytes_feature(str(tmp[col]).encode(encoding="utf-8"))
+            features[col] = _bytes_feature(tmp[col])
         example = tf.train.Example(features=tf.train.Features(feature=features))
         examples.append(example.SerializeToString())
     total_1 = (timeit.default_timer() - time_1)
     print("make example cost {:.5f}s".format(total_1))
+    return examples
+
+
+def model_predict2(device_id, diary_ids, device_dict, diary_dict, predict_fn):
+    time_1 = timeit.default_timer()
+    device_info, diary_lst = device_diary_fe(device_id, diary_ids, device_dict, diary_dict)
+
+    examples = _make_example(device_info, diary_lst)
 
     time_1 = timeit.default_timer()
     predictions = predict_fn({"examples": examples})