test

eda/esmm/Model_pipline/dist_predict.py

@@ -142,7 +142,7 @@ def input_fn(filenames, batch_size=32, num_epochs=1, perform_shuffle=False):
     #print(batch_features,batch_labels)
     return batch_features, batch_labels
 
-def main(_):
+def main(te_file):
     dt_dir = (date.today() + timedelta(-1)).strftime('%Y%m%d')
     model_dir = "hdfs://172.16.32.4:8020/strategy/esmm/model_ckpt/DeepCvrMTL/" + dt_dir
     te_files = ["hdfs://172.16.32.4:8020/strategy/esmm/nearby/part-r-00000"]
@@ -160,7 +160,7 @@ def main(_):
             log_step_count_steps=100, save_summary_steps=100)
     Estimator = tf.estimator.Estimator(model_fn=model_fn, model_dir=model_dir, params=model_params, config=config)
 
-    preds = Estimator.predict(input_fn=lambda: input_fn(te_files, num_epochs=1, batch_size=10000), predict_keys=["pctcvr","pctr","pcvr"])
+    preds = Estimator.predict(input_fn=lambda: input_fn(te_file, num_epochs=1, batch_size=10000), predict_keys=["pctcvr","pctr","pcvr"])
 
     with open("/home/gmuser/esmm/nearby/pred.txt", "w") as fo:
         for prob in preds:
@@ -202,8 +202,8 @@ if __name__ == "__main__":
     # rdd_te_files = spark.sparkContext.parallelize(te_files)
     # indices = rdd_te_files.repartition(2).map(lambda x: main(x))
     # print(indices.collect())
-    tf.app.run()
-
+    # tf.app.run()
+    main(te_files)
 
     b = time.time()
     print("耗时(分钟)：")