separate logger

Author: js1010

cpp/include/culda/culda.hpp

@@ -75,6 +75,7 @@ class CuLDA {
   DeviceInfo dev_info_;
   json11::Json opt_;
   std::shared_ptr<spdlog::logger> logger_;
+  std::unique_ptr<CuSimLogger> logger_container_;
   thrust::device_vector<float> dev_alpha_, dev_beta_;
   thrust::device_vector<float> dev_grad_alpha_, dev_new_beta_;
   thrust::device_vector<float> dev_gamma_, dev_new_gamma_, dev_phi_;

cpp/include/cuw2v/cuw2v.hpp

@@ -52,6 +52,7 @@ class CuW2V {
   DeviceInfo dev_info_;
   json11::Json opt_;
   std::shared_ptr<spdlog::logger> logger_;
+  std::unique_ptr<CuSimLogger> logger_container_;
   int block_cnt_, block_dim_;
   int num_dims_, num_words_, window_size_;
   float *emb_in_, *emb_out_, lr_;

cpp/include/utils/ioutils.hpp

@@ -45,6 +45,7 @@ class IoUtils {
   std::ifstream stream_fin_;
   json11::Json opt_;
   std::shared_ptr<spdlog::logger> logger_;
+  std::unique_ptr<CuSimLogger> logger_container_;
   std::unordered_map<std::string, int> word_idmap_, word_count_;
   std::vector<std::string> word_list_;
   int num_lines_, remain_lines_;

cpp/include/utils/log.hpp

@@ -7,7 +7,7 @@
 // reference: https://github.com/kakao/buffalo/blob/5f571c2c7d8227e6625c6e538da929e4db11b66d/lib/misc/log.cc
 #pragma once
 #include <memory>
-
+#include <string>
 #define SPDLOG_EOL ""
 #define SPDLOG_TRACE_ON
 #include "spdlog/spdlog.h"
@@ -32,6 +32,7 @@ namespace cusim {
 class CuSimLogger {
  public:
   CuSimLogger();
+  explicit CuSimLogger(std::string name);
   std::shared_ptr<spdlog::logger>& get_logger();
   void set_log_level(int level);
   int get_log_level();

cpp/src/culda/culda.cu

@@ -9,7 +9,8 @@
 namespace cusim {
 
 CuLDA::CuLDA() {
-  logger_ = CuSimLogger().get_logger();
+  logger_container_.reset(new CuSimLogger("lda"));
+  logger_ = logger_container_->get_logger();
   dev_info_ = GetDeviceInfo();
   if (dev_info_.unknown) DEBUG0("Unknown device type");
   INFO("cuda device info, major: {}, minor: {}, multi processors: {}, cores: {}",
@@ -28,7 +29,7 @@ bool CuLDA::Init(std::string opt_path) {
   auto _opt = json11::Json::parse(str, err_cmt);
   if (not err_cmt.empty()) return false;
   opt_ = _opt;
-  CuSimLogger().set_log_level(opt_["c_log_level"].int_value());
+  logger_container_->set_log_level(opt_["c_log_level"].int_value());
   num_topics_ = opt_["num_topics"].int_value();
   block_dim_ = opt_["block_dim"].int_value();
   block_cnt_ = opt_["hyper_threads"].number_value() * (dev_info_.cores / block_dim_);

cpp/src/cuw2v/cuw2v.cu

@@ -24,7 +24,8 @@ bool CompareIndex(int lhs, int rhs) {
 }
 
 CuW2V::CuW2V() {
-  logger_ = CuSimLogger().get_logger();
+  logger_container_.reset(new CuSimLogger("w2v"));
+  logger_ = logger_container_->get_logger();
   dev_info_ = GetDeviceInfo();
   if (dev_info_.unknown) DEBUG0("Unknown device type");
   INFO("cuda device info, major: {}, minor: {}, multi processors: {}, cores: {}",
@@ -43,7 +44,7 @@ bool CuW2V::Init(std::string opt_path) {
   auto _opt = json11::Json::parse(str, err_cmt);
   if (not err_cmt.empty()) return false;
   opt_ = _opt;
-  CuSimLogger().set_log_level(opt_["c_log_level"].int_value());
+  logger_container_->set_log_level(opt_["c_log_level"].int_value());
   num_dims_ = opt_["num_dims"].int_value();
   block_dim_ = opt_["block_dim"].int_value();
   block_cnt_ = opt_["hyper_threads"].number_value() * (dev_info_.cores / block_dim_);

cpp/src/utils/ioutils.cc

@@ -8,7 +8,8 @@
 namespace cusim {
 
 IoUtils::IoUtils() {
-  logger_ = CuSimLogger().get_logger();
+  logger_container_.reset(new CuSimLogger("ioutils"));
+  logger_ = logger_container_->get_logger();
 }
 
 IoUtils::~IoUtils() {}
@@ -23,7 +24,7 @@ bool IoUtils::Init(std::string opt_path) {
   auto _opt = json11::Json::parse(str, err_cmt);
   if (not err_cmt.empty()) return false;
   opt_ = _opt;
-  CuSimLogger().set_log_level(opt_["c_log_level"].int_value());
+  logger_container_->set_log_level(opt_["c_log_level"].int_value());
   return true;
 }
 

cpp/src/utils/log.cc

@@ -16,18 +16,26 @@ CuSimLogger::CuSimLogger() {
   logger_ = spdlog::default_logger();
 }
 
+CuSimLogger::CuSimLogger(std::string name) {
+  // auto console_sink = std::make_shared<spdlog::sinks::stdout_color_sink_mt>();
+  auto stderr_sink = std::make_shared<spdlog::sinks::stderr_color_sink_mt>();
+  // spdlog::sinks_init_list sinks = {console_sink, stderr_sink};
+  logger_ = std::make_shared<spdlog::logger>(name, stderr_sink);
+  logger_->set_pattern("[%^%-8l%$] %Y-%m-%d %H:%M:%S %v");
+}
+
 std::shared_ptr<spdlog::logger>& CuSimLogger::get_logger() {
   return logger_;
 }
 
 void CuSimLogger::set_log_level(int level) {
   global_logging_level_ = level;
   switch (level) {
-    case 0: spdlog::set_level(spdlog::level::off); break;
-    case 1: spdlog::set_level(spdlog::level::warn); break;
-    case 2: spdlog::set_level(spdlog::level::info); break;
-    case 3: spdlog::set_level(spdlog::level::debug); break;
-    default: spdlog::set_level(spdlog::level::trace); break;
+    case 0: logger_->set_level(spdlog::level::off); break;
+    case 1: logger_->set_level(spdlog::level::warn); break;
+    case 2: logger_->set_level(spdlog::level::info); break;
+    case 3: logger_->set_level(spdlog::level::debug); break;
+    default: logger_->set_level(spdlog::level::trace); break;
   }
 }
 

cusim/cuw2v/pycuw2v.py

@@ -86,7 +86,10 @@ def init_model(self):
   def train_model(self):
     self.preprocess_data()
     self.init_model()
-    if not self.opt.neg:
+    if self.opt.neg:
+      self.obj.build_random_table( \
+        self.word_count, self.opt.random_size, self.opt.num_threads)
+    else:
       self.obj.build_huffman_tree(self.word_count)
     h5f = h5py.File(pjoin(self.opt.processed_data_dir, "token.h5"), "r")
     for epoch in range(1, self.opt.epochs + 1):
@@ -112,9 +115,6 @@ def _train_epoch(self, h5f):
       offset = next_offset
 
       # call cuda kernel
-      if self.opt.neg:
-        self.obj.build_random_table( \
-          self.word_count, self.opt.random_size, self.opt.num_threads)
       _loss_nume, _loss_deno = \
         self.obj.feed_data(cols, indptr)
 

cusim/proto/config.proto

@@ -51,7 +51,7 @@ message CuW2VConfigProto {
   optional int32 seed = 14 [default = 777];
   optional int32 table_seed = 15 [default = 777];
   optional int32 cuda_seed = 16 [default = 777];
-  optional int32 random_size = 17 [default = 1000000];
+  optional int32 random_size = 12 [default = 1000000];
 
   optional int32 neg = 17 [default = 10];
   // as recommended in w2v paper