Merge pull request #1451 from jacquesqiao/fix-ctr

fix ctr reader

fluid/PaddleRec/ctr/network_conf.py

@@ -15,8 +15,12 @@ def ctr_dnn_model(embedding_size, sparse_feature_dim):
     def embedding_layer(input):
         return fluid.layers.embedding(
             input=input,
+            is_sparse=True,
+            # you need to patch https://github.com/PaddlePaddle/Paddle/pull/14190
+            # if you want to set is_distributed to True
+            is_distributed=False,
             size=[sparse_feature_dim, embedding_size],
-            param_attr=fluid.ParamAttr(name="SparseFeatFactors", initializer=fluid.initializer.Normal(scale=1/math.sqrt(sparse_feature_dim))))
+            param_attr=fluid.ParamAttr(name="SparseFeatFactors", initializer=fluid.initializer.Uniform()))
 
     sparse_embed_seq = map(embedding_layer, sparse_input_ids)
     concated = fluid.layers.concat(sparse_embed_seq + [dense_input], axis=1)

fluid/PaddleRec/ctr/reader.py

@@ -21,10 +21,10 @@ class CriteoDataset(Dataset):
                     for line in f:
                         line_idx += 1
                         if is_train and line_idx > self.train_idx_:
-                            continue
+                            break
                         elif not is_train and line_idx <= self.train_idx_:
                             continue
-                        if trainer_id > 0 and line_idx % trainer_num != trainer_id:
+                        if line_idx % trainer_num != trainer_id:
                             continue
                         features = line.rstrip('\n').split('\t')
                         dense_feature = []

fluid/PaddleRec/ctr/train.py

@@ -138,7 +138,7 @@ def train():
     if args.is_local:
         logger.info("run local training")
         main_program = fluid.default_main_program()
-        train_loop(args, main_program, data_list, loss, auc_var, batch_auc_var, 1, -1)
+        train_loop(args, main_program, data_list, loss, auc_var, batch_auc_var, 1, 0)
     else:
         logger.info("run dist training")
         t = fluid.DistributeTranspiler()
@@ -154,7 +154,7 @@ def train():
             logger.info("run trainer")
             train_prog = t.get_trainer_program()
             train_loop(args, train_prog, data_list, loss, auc_var, batch_auc_var, 
-                       args.trainers, args.trainer_id + 1)
+                       args.trainers, args.trainer_id)
 
 
 if __name__ == '__main__':