split file

tensnsorflow/test.py

@@ -26,7 +26,6 @@ def gen_tfrecords(in_file):
     out_file = os.path.join(FLAGS.output_dir, basename)
     tfrecord_out = tf.python_io.TFRecordWriter(out_file)
     df = pd.read_csv(in_file)
-    print(df.head(2))
 
     for i in range(df.shape[0]):
         features = tf.train.Features(feature={

tensnsorflow/train.py

@@ -109,13 +109,18 @@ def model_fn(features, labels, mode, params):
 
     #------build feaure-------
     #{U-A-X-C不需要特殊处理的特征}
-    feat_ids    = features['feat_ids']
-    #feat_vals   = features['feat_vals']
-    #{User multi-hot}
-    #{Ad}
-    #{X multi-hot}
-    #x_intids    = features['x_intids']
-    #x_intvals   = features['x_intvals']
+
+    ucity_id    = features['ucity_id']
+    clevel1_id = features['clevel1_id']
+    ccity_name = features['ccity_name']
+    device_type = features['device_type']
+    manufacturer = features['manufacturer']
+    channel = features['channel']
+    top = features['top']
+    level2_ids = features['level2_ids']
+    time = features['time']
+    stat_date = features['stat_date']
+
 
     if FLAGS.task_type != "infer":
         y = labels['y']
@@ -123,10 +128,20 @@ def model_fn(features, labels, mode, params):
 
     #------build f(x)------
     with tf.variable_scope("Shared-Embedding-layer"):
-        common_embs = tf.nn.embedding_lookup(Feat_Emb, feat_ids)     # None * F' * K
-        #common_embs = tf.multiply(common_embs, feat_vals)
-
-        x_concat = tf.concat([tf.reshape(common_embs,shape=[-1, common_dims])],axis=1)    # None * (F * K)
+        ucity_id = tf.nn.embedding_lookup(Feat_Emb, ucity_id)
+        clevel1_id = tf.nn.embedding_lookup(Feat_Emb, clevel1_id)
+        ccity_name = tf.nn.embedding_lookup(Feat_Emb, ccity_name)
+        device_type = tf.nn.embedding_lookup(Feat_Emb, device_type)
+        manufacturer = tf.nn.embedding_lookup(Feat_Emb, manufacturer)
+        channel = tf.nn.embedding_lookup(Feat_Emb, channel)
+        top = tf.nn.embedding_lookup(Feat_Emb, top)
+        level2_ids = tf.nn.embedding_lookup(Feat_Emb, level2_ids)
+        time = tf.nn.embedding_lookup(Feat_Emb, time)
+        stat_date = tf.nn.embedding_lookup(Feat_Emb, stat_date)
+
+
+        x_concat = tf.concat([ucity_id,clevel1_id,ccity_name,device_type,manufacturer,
+                              channel,top,level2_ids,time,stat_date],axis=1)    # None * (F * K)
 
     with tf.name_scope("CVR_Task"):
         if mode == tf.estimator.ModeKeys.TRAIN: