add is_sparse back (#2729)

docs/Parameters.rst

@@ -501,6 +501,10 @@ IO Parameters
 
    -  LightGBM will auto compress memory according to ``max_bin``. For example, LightGBM will use ``uint8_t`` for feature value if ``max_bin=255``
 
+-  ``is_enable_sparse`` :raw-html:`<a id="is_enable_sparse" title="Permalink to this parameter" href="#is_enable_sparse">&#x1F517;&#xFE0E;</a>`, default = ``true``, type = bool, aliases: ``is_sparse``, ``enable_sparse``, ``sparse``
+
+   -  used to enable/disable sparse optimization
+
 -  ``max_bin_by_feature`` :raw-html:`<a id="max_bin_by_feature" title="Permalink to this parameter" href="#max_bin_by_feature">&#x1F517;&#xFE0E;</a>`, default = ``None``, type = multi-int
 
    -  max number of bins for each feature

include/LightGBM/config.h

@@ -478,6 +478,11 @@ struct Config {
   // desc = LightGBM will auto compress memory according to ``max_bin``. For example, LightGBM will use ``uint8_t`` for feature value if ``max_bin=255``
   int max_bin = 255;
 
+
+  // alias = is_sparse, enable_sparse, sparse
+  // desc = used to enable/disable sparse optimization
+  bool is_enable_sparse = true;
+
   // type = multi-int
   // default = None
   // desc = max number of bins for each feature

src/io/config_auto.cpp

@@ -92,6 +92,9 @@ std::unordered_map<std::string, std::string> Config::alias_table({
   {"forced_splits_file", "forcedsplits_filename"},
   {"forced_splits", "forcedsplits_filename"},
   {"verbose", "verbosity"},
+  {"is_sparse", "is_enable_sparse"},
+  {"enable_sparse", "is_enable_sparse"},
+  {"sparse", "is_enable_sparse"},
   {"subsample_for_bin", "bin_construct_sample_cnt"},
   {"hist_pool_size", "histogram_pool_size"},
   {"data_seed", "data_random_seed"},
@@ -222,6 +225,7 @@ std::unordered_set<std::string> Config::parameter_set({
   "cegb_penalty_feature_coupled",
   "verbosity",
   "max_bin",
+  "is_enable_sparse",
   "max_bin_by_feature",
   "min_data_in_bin",
   "bin_construct_sample_cnt",
@@ -435,6 +439,8 @@ void Config::GetMembersFromString(const std::unordered_map<std::string, std::str
   GetInt(params, "max_bin", &max_bin);
   CHECK(max_bin >1);
 
+  GetBool(params, "is_enable_sparse", &is_enable_sparse);
+
   if (GetString(params, "max_bin_by_feature", &tmp_str)) {
     max_bin_by_feature = Common::StringToArray<int32_t>(tmp_str, ',');
   }
@@ -634,6 +640,7 @@ std::string Config::SaveMembersToString() const {
   str_buf << "[cegb_penalty_feature_coupled: " << Common::Join(cegb_penalty_feature_coupled, ",") << "]\n";
   str_buf << "[verbosity: " << verbosity << "]\n";
   str_buf << "[max_bin: " << max_bin << "]\n";
+  str_buf << "[is_enable_sparse: " << is_enable_sparse << "]\n";
   str_buf << "[max_bin_by_feature: " << Common::Join(max_bin_by_feature, ",") << "]\n";
   str_buf << "[min_data_in_bin: " << min_data_in_bin << "]\n";
   str_buf << "[bin_construct_sample_cnt: " << bin_construct_sample_cnt << "]\n";

src/io/dataset.cpp

@@ -98,6 +98,7 @@ std::vector<std::vector<int>> FindGroups(const std::vector<std::unique_ptr<BinMa
                                          data_size_t total_sample_cnt,
                                          data_size_t num_data,
                                          bool is_use_gpu,
+                                         bool is_sparse,
                                          std::vector<int8_t>* multi_val_group) {
   const int max_search_group = 100;
   const int max_bin_per_group = 256;
@@ -165,6 +166,10 @@ std::vector<std::vector<int>> FindGroups(const std::vector<std::unique_ptr<BinMa
       group_num_bin.push_back(1 + bin_mappers[fidx]->num_bin() + (bin_mappers[fidx]->GetDefaultBin() == 0 ? -1 : 0));
     }
   }
+  if (!is_sparse) {
+    multi_val_group->resize(features_in_group.size(), false);
+    return features_in_group;
+  }
   std::vector<int> second_round_features;
   std::vector<std::vector<int>> features_in_group2;
   std::vector<std::vector<bool>> conflict_marks2;
@@ -217,6 +222,7 @@ std::vector<std::vector<int>> FastFeatureBundling(const std::vector<std::unique_
                                                   const std::vector<int>& used_features,
                                                   data_size_t num_data,
                                                   bool is_use_gpu,
+                                                  bool is_sparse,
                                                   std::vector<int8_t>* multi_val_group) {
   Common::FunctionTimer fun_timer("Dataset::FastFeatureBundling", global_timer);
   std::vector<size_t> feature_non_zero_cnt;
@@ -263,8 +269,8 @@ std::vector<std::vector<int>> FastFeatureBundling(const std::vector<std::unique_
     }
   }
   std::vector<int8_t> group_is_multi_val, group_is_multi_val2;
-  auto features_in_group = FindGroups(bin_mappers, used_features, sample_indices, tmp_num_per_col.data(), num_sample_col, total_sample_cnt, num_data, is_use_gpu, &group_is_multi_val);
-  auto group2 = FindGroups(bin_mappers, feature_order_by_cnt, sample_indices, tmp_num_per_col.data(), num_sample_col, total_sample_cnt, num_data, is_use_gpu, &group_is_multi_val2);
+  auto features_in_group = FindGroups(bin_mappers, used_features, sample_indices, tmp_num_per_col.data(), num_sample_col, total_sample_cnt, num_data, is_use_gpu, is_sparse, &group_is_multi_val);
+  auto group2 = FindGroups(bin_mappers, feature_order_by_cnt, sample_indices, tmp_num_per_col.data(), num_sample_col, total_sample_cnt, num_data, is_use_gpu, is_sparse, &group_is_multi_val2);
 
   if (features_in_group.size() > group2.size()) {
     features_in_group = group2;
@@ -311,7 +317,7 @@ void Dataset::Construct(
   if (io_config.enable_bundle && !used_features.empty()) {
     features_in_group = FastFeatureBundling(*bin_mappers,
                                             sample_non_zero_indices, sample_values, num_per_col, num_sample_col, static_cast<data_size_t>(total_sample_cnt),
-                                            used_features, num_data_, io_config.device_type == std::string("gpu"), &group_is_multi_val);
+                                            used_features, num_data_, io_config.device_type == std::string("gpu"), io_config.is_enable_sparse, &group_is_multi_val);
   }
 
   num_features_ = 0;