[ModelZoo] unify optimizer loss names

Dockerfile

@@ -89,7 +89,7 @@ ARG WITH_SQLFLOW_MODELS="ON"
 RUN if [ "${WITH_SQLFLOW_MODELS:-ON}" = "ON" ]; then \
   git clone https://github.com/sql-machine-learning/models.git && \
   cd models && \
-  git checkout 58f4c137129e2bc749320bafcc8fddb7c737fed9 && \
+  git checkout 91d63b581fad5686a2132635f0aa000b0699a1da && \
   bash -c "source activate sqlflow-dev && python setup.py install" && \
   cd .. && \
   rm -rf models; \

python/sqlflow_submitter/tensorflow/predict.py

@@ -83,6 +83,7 @@ def pred(is_keras_model,
         classifier = estimator(**feature_columns, **model_params, model_dir=save)
     else:
         classifier = estimator(**feature_columns, **model_params)
+        classifier_pkg = sys.modules[estimator.__module__]
 
 
     if is_keras_model:
@@ -123,7 +124,7 @@ def eval_input_fn(batch_size):
                 except tf.errors.OutOfRangeError:
                     break
                 result = classifier.predict_on_batch(features[0])
-                result = classifier.prepare_prediction_column(result[0])
+                result = classifier_pkg.prepare_prediction_column(result[0])
                 row = []
                 for idx, name in enumerate(feature_column_names):
                     val = features[0][name].numpy()[0]

python/sqlflow_submitter/tensorflow/train.py

@@ -18,6 +18,7 @@
 import sys, json
 import tensorflow as tf
 import functools
+import sys
 try:
     import sqlflow_models
 except:
@@ -68,6 +69,7 @@ def train(is_keras_model,
         classifier = estimator(**feature_columns, **model_params, model_dir=save)
     else:
         classifier = estimator(**feature_columns, **model_params)
+        classifier_pkg = sys.modules[estimator.__module__]
 
     def input_fn(datasetStr):
         feature_types = []
@@ -96,8 +98,8 @@ def validate_input_fn(batch_size):
         return dataset.batch(batch_size)
 
     if is_keras_model:
-        classifier.compile(optimizer=classifier.default_optimizer(),
-            loss=classifier.default_loss(),
+        classifier.compile(optimizer=classifier_pkg.optimizer(),
+            loss=classifier_pkg.loss(),
             metrics=["accuracy"])
         if hasattr(classifier, 'sqlflow_train_loop'):
             classifier.sqlflow_train_loop(train_input_fn(batch_size))