fix criteo ctr with cube

core/general-server/op/general_dist_kv_infer_op.cpp

@@ -186,9 +186,9 @@ int GeneralDistKVInferOp::inference() {
   if (values.size() != keys.size() || values[0].buff.size() == 0) {
     LOG(ERROR) << "cube value return null";
   }
-  // size_t EMBEDDING_SIZE = values[0].buff.size() / sizeof(float);
+   size_t EMBEDDING_SIZE = values[0].buff.size() / sizeof(float);
   // size_t EMBEDDING_SIZE = (values[0].buff.size() - 10) / sizeof(float);
-  size_t EMBEDDING_SIZE = 9;
+  //size_t EMBEDDING_SIZE = 9;
   TensorVector sparse_out;
   sparse_out.resize(sparse_count);
   TensorVector dense_out;
@@ -241,7 +241,7 @@ int GeneralDistKVInferOp::inference() {
 
       // The data generated by pslib has 10 bytes of information to be filtered
       // out
-      memcpy(data_ptr, cur_val->buff.data() + 10, cur_val->buff.size() - 10);
+      memcpy(data_ptr, cur_val->buff.data(), cur_val->buff.size() );
       // VLOG(3) <<  keys[cube_val_idx] << ":" << data_ptr[0] << ", " <<
       // data_ptr[1] << ", " <<data_ptr[2] << ", " <<data_ptr[3] << ", "
       // <<data_ptr[4] << ", " <<data_ptr[5] << ", " <<data_ptr[6] << ", "

python/examples/criteo_ctr_with_cube/test_client.py

@@ -35,7 +35,7 @@ reader = dataset.infer_reader(test_filelists, batch, buf_size)
 label_list = []
 prob_list = []
 start = time.time()
-for ei in range(1):
+for ei in range(100):
     if py_version == 2:
         data = reader().next()
     else:
@@ -46,16 +46,12 @@ for ei in range(1):
     for i in range(1, 27):
         feed_dict["embedding_{}.tmp_0".format(i - 1)] = np.array(data[0][i]).reshape(len(data[0][i]))
         feed_dict["embedding_{}.tmp_0.lod".format(i - 1)] = [0, len(data[0][i])]
-    for key in feed_dict.keys():
-        if "lod" not in key:
-            print("key: {}, shape: {}".format(key, feed_dict[key].shape))
     fetch_map = client.predict(feed=feed_dict, fetch=["prob"],batch=True)
     print(fetch_map)
     prob_list.append(fetch_map['prob'][0][1])
     label_list.append(data[0][-1][0])
 
 
-print(auc(label_list, prob_list))
 end = time.time()
 print(end - start)