模型上线

train/train_service_sk_tf2_distibute.py

 import tensorflow.compat.v1 as tf
-tf.disable_eager_execution()
 tf.logging.set_verbosity(tf.logging.INFO)
 import sys
 import time
@@ -220,8 +219,7 @@ session_config = tf.compat.v1.ConfigProto(allow_soft_placement = True)
 
 session_config.gpu_options.allow_growth = True
 
-# config = tf.estimator.RunConfig(save_checkpoints_steps = 3000, train_distribute = distribution, eval_distribute = distribution)
-config = tf.estimator.RunConfig(save_checkpoints_steps = 300, train_distribute = distribution, eval_distribute = distribution)
+config = tf.estimator.RunConfig(save_checkpoints_steps = 3000, train_distribute = distribution, eval_distribute = distribution)
 
 # config = tf.estimator.RunConfig(save_checkpoints_steps = 3000, session_config = session_config)
 
@@ -238,17 +236,7 @@ early_stopping = tf.estimator.experimental.stop_if_no_increase_hook(wideAndDeepM
 
 hooks = [early_stopping]
 
-train_dataset = input_fn(DATA_DIR + 'train_samples.csv', 1, True, 2048)
-
-eval_dataset = input_fn(DATA_DIR + 'eval_samples.csv', 1, False, 2 ** 15)
-
-def train_dataset_input_fn(train_dataset):
-    return train_dataset
-
-def eval_dataset_input_fn(eval_dataset):
-    return eval_dataset
-
-train_spec = tf.estimator.TrainSpec(input_fn = lambda: input_fn(DATA_DIR + 'train_samples.csv', 1, True, 2048), hooks = hooks)
+train_spec = tf.estimator.TrainSpec(input_fn = lambda: input_fn(DATA_DIR + 'train_samples.csv', 3, True, 2048), hooks = hooks)
 
 serving_feature_spec = tf.feature_column.make_parse_example_spec(
   linear_feature_columns + dnn_feature_columns)
@@ -262,7 +250,7 @@ exporter = tf.estimator.BestExporter(
   serving_input_receiver_fn = serving_input_receiver_fn,
   exports_to_keep = 1)
 
-eval_spec = tf.estimator.EvalSpec(input_fn = lambda: eval_dataset_input_fn(eval_dataset), steps = None, throttle_secs = 120, exporters = exporter)
+eval_spec = tf.estimator.EvalSpec(input_fn = lambda: input_fn(DATA_DIR + 'eval_samples.csv', 1, False, 2 ** 15), steps = None, throttle_secs = 120, exporters = exporter)
 
 # def my_auc(labels, predictions):
 #     return {'auc_pr_careful_interpolation': tf.metrics.auc(labels, predictions['logistic'], curve='ROC',