change slot_num and layer_size

Author: esythan

models/rank/slot_dnn/config_offline_infer.yaml

@@ -28,13 +28,14 @@ runner:
   reader_type: "InmemoryDataset"  # DataLoader / QueueDataset / RecDataset
   pipe_command: "python3 inmemorydataset_reader.py"
 
-  init_model_path: "output_model_benchdnn_queue/20190720/6"
+  init_model_path: "output_model/20190720/inference_model_6"
 
   dataset_debug: False
   parse_ins_id: True
   dump_fields_path: "dump_offline_infer"
 
   shots_shuffle_list: [["2"], ["3"]]
+  candidate_size: 10
 
 
 # hyper parameters of user-defined network
@@ -48,6 +49,6 @@ hyper_parameters:
   dict_dim : 1000000
   emb_dim : 9
   sparse_feature_dim: 9
-  slot_num: 408
-  layer_sizes: [512, 256, 128, 128, 128]
+  slot_num: 300
+  layer_sizes: [512, 256, 128]
   distributed_embedding: 0

models/rank/slot_dnn/config_online.yaml

@@ -1,7 +1,7 @@
 runner:
   use_gloo: True
   # train_data_dir: "afs:/xxx"
-  train_data_dir: "./data2"
+  train_data_dir: "./data"
   train_reader_path: "criteo_reader" # importlib format
   use_gpu: False
   use_auc: True
@@ -54,6 +54,6 @@ hyper_parameters:
   dict_dim : 1000000
   emb_dim : 9
   sparse_feature_dim: 9
-  slot_num: 408
-  layer_sizes: [512, 256, 128, 128, 128]
+  slot_num: 300
+  layer_sizes: [512, 256, 128]
   distributed_embedding: 0

models/rank/slot_dnn/data/demo_10

@@ -32,7 +32,7 @@ def init(self, config):
         self.config = config
         padding = "0"
         #sparse_slots = "click 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26"
-        self.slots = self.slot_reader(slot_num=408)
+        self.slots = self.slot_reader(slot_num=300)
         self.slot2index = {}
         self.visit = {}
         for i in range(len(self.slots)):
@@ -66,7 +66,7 @@ def line_process(self, line):
             slot = slot_feasign[1]
             if slot not in self.slots:
                 continue
-            feasign = int(slot_feasign[0])
+            feasign = slot_feasign[0]
             # if feasign not in fea_dict:
             #     fea_dict[feasign] = str(len(fea_dict))
             # output[self.slot2index[slot]][1].append(fea_dict[feasign])

models/rank/slot_dnn/inmemorydataset_reader.py

@@ -68,7 +68,6 @@ def forward(self, slot_inputs):
         self.inference_feed_vars = []
         for s_input in slot_inputs:
             emb = paddle.static.nn.sparse_embedding(
-                #emb = paddle.fluid.contrib.sparse_embedding(
                 input=s_input,
                 size=[self.dict_dim, self.emb_dim],
                 param_attr=paddle.ParamAttr(name="embedding"))

models/rank/slot_dnn/net.py

@@ -32,7 +32,7 @@ def init(self, config):
         self.config = config
         padding = 0
         #sparse_slots = "click 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26"
-        self.slots = self.slot_reader(slot_num=408)
+        self.slots = self.slot_reader(slot_num=300)
         self.slot2index = {}
         self.visit = {}
         for i in range(len(self.slots)):

models/rank/slot_dnn/online_learning_config.yaml

@@ -134,7 +134,7 @@ def run_offline_infer(self):
 
         self.exe.run(paddle.static.default_startup_program())
         fleet.init_worker()
-        fleet.load_model(init_model_path, mode="0")
+        fleet.load_model(init_model_path, mode=0)
 
         logger.info("Prepare Dataset Begin.")
         prepare_data_start_time = time.time()
@@ -169,8 +169,9 @@ def dataset_offline_infer(self, cur_dataset):
                       self.metric_types)
         logger.info("baseline auc: {}".format(baseline_auc))
         slots_shuffle_list = config.get("runner.shots_shuffle_list", [])
+        candidate_size = config.get("runner.candidate_size", 10)
         for slots_list in slots_shuffle_list:
-            cur_dataset.set_fea_eval(10, True)
+            cur_dataset.set_fea_eval(candidate_size, True)
             cur_dataset.slots_shuffle(slots_list)
             self.exe.infer_from_dataset(
                 program=paddle.static.default_main_program(),