Merge pull request #5 from google/test_374585820

Release of v0.1.3
google · May 19, 2021 · e2c9a49 · e2c9a49
2 parents 4903e46 + 31932e6
commit e2c9a49
Show file tree

Hide file tree

Showing 3 changed files with 304 additions and 2 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,11 @@
 # Changelog
 
+## 0.1.3 - 2021-05-19
+
+### Features
+
+-   Register new inference engines.
+
 ## 0.1.2 - 2021-05-18
 
 ### Features

diff --git a/README.md b/README.md
@@ -95,8 +95,8 @@ Download one of the build, and then run `examples/beginner.{sh,bat}`.
 
 Target  | Version | Link
 ------- | ------- | ----
-Linux   | 0.1.2   | [CLI](https://github.com/google/yggdrasil-decision-forests/releases/download/0.1.2/cli_linux.zip)
-Windows | 0.1.2   | [CLI](https://github.com/google/yggdrasil-decision-forests/releases/download/0.1.2/cli_windows.zip)
+Linux   | 0.1.0   | [CLI](https://github.com/google/yggdrasil-decision-forests/releases/download/0.1.0/cli_linux.zip)
+Windows | 0.1.0   | [CLI](https://github.com/google/yggdrasil-decision-forests/releases/download/0.1.0/cli_windows.zip)
 
 ## Installation from Source
 

diff --git a/yggdrasil_decision_forests/serving/decision_forest/register_engines.cc b/yggdrasil_decision_forests/serving/decision_forest/register_engines.cc
@@ -21,6 +21,7 @@
 #include "yggdrasil_decision_forests/model/fast_engine_factory.h"
 #include "yggdrasil_decision_forests/model/gradient_boosted_trees/gradient_boosted_trees.h"
 #include "yggdrasil_decision_forests/serving/decision_forest/decision_forest.h"
+#include "yggdrasil_decision_forests/serving/decision_forest/quick_scorer_extended.h"
 #include "yggdrasil_decision_forests/serving/example_set_model_wrapper.h"
 #include "yggdrasil_decision_forests/utils/compatibility.h"
 
@@ -130,6 +131,212 @@ class GradientBoostedTreesGenericFastEngineFactory : public FastEngineFactory {
 REGISTER_FastEngineFactory(GradientBoostedTreesGenericFastEngineFactory,
                            serving::gradient_boosted_trees::kGeneric);
 
+class GradientBoostedTreesQuickScorerFastEngineFactory
+    : public FastEngineFactory {
+ public:
+  using SourceModel = gradient_boosted_trees::GradientBoostedTreesModel;
+
+  std::string name() const override {
+    return serving::gradient_boosted_trees::kQuickScorerExtended;
+  }
+
+  bool IsCompatible(const AbstractModel* const model) const override {
+    auto* gbt_model = dynamic_cast<const SourceModel*>(model);
+    if (gbt_model == nullptr) {
+      return false;
+    }
+
+    if (!gbt_model->IsMissingValueConditionResultFollowGlobalImputation()) {
+      return false;
+    }
+
+    if (gbt_model->NumTrees() > serving::decision_forest::internal::
+                                    QuickScorerExtendedModel::kMaxTrees) {
+      return false;
+    }
+
+    for (const auto& src_tree : gbt_model->decision_trees()) {
+      if (src_tree->NumLeafs() > serving::decision_forest::internal::
+                                     QuickScorerExtendedModel::kMaxLeafs) {
+        return false;
+      }
+    }
+
+    switch (gbt_model->task()) {
+      case proto::CLASSIFICATION:
+        return gbt_model->label_col_spec()
+                   .categorical()
+                   .number_of_unique_values() == 3;
+      case proto::REGRESSION:
+      case proto::RANKING:
+        return true;
+      default:
+        return false;
+    }
+  }
+
+  std::vector<std::string> IsBetterThan() const override {
+    return {serving::gradient_boosted_trees::kGeneric,
+            serving::gradient_boosted_trees::kOptPred};
+  }
+
+  utils::StatusOr<std::unique_ptr<serving::FastEngine>> CreateEngine(
+      const AbstractModel* const model) const override {
+    auto* gbt_model = dynamic_cast<const SourceModel*>(model);
+    if (!gbt_model) {
+      return absl::InvalidArgumentError("The model is not a GBDT.");
+    }
+
+    switch (gbt_model->task()) {
+      case proto::CLASSIFICATION:
+        if (gbt_model->label_col_spec()
+                .categorical()
+                .number_of_unique_values() == 3) {
+          // Binary classification.
+          auto engine = absl::make_unique<serving::ExampleSetModelWrapper<
+              serving::decision_forest::
+                  GradientBoostedTreesBinaryClassificationQuickScorerExtended,
+              serving::decision_forest::Predict>>();
+          RETURN_IF_ERROR(engine->LoadModel<SourceModel>(*gbt_model));
+          return engine;
+        } else {
+          return absl::InvalidArgumentError("Non supported GBDT model");
+        }
+
+      case proto::REGRESSION: {
+        auto engine = absl::make_unique<serving::ExampleSetModelWrapper<
+            serving::decision_forest::
+                GradientBoostedTreesRegressionQuickScorerExtended,
+            serving::decision_forest::Predict>>();
+        RETURN_IF_ERROR(engine->LoadModel<SourceModel>(*gbt_model));
+        return engine;
+      }
+
+      case proto::RANKING: {
+        auto engine = absl::make_unique<serving::ExampleSetModelWrapper<
+            serving::decision_forest::
+                GradientBoostedTreesRankingQuickScorerExtended,
+            serving::decision_forest::Predict>>();
+        RETURN_IF_ERROR(engine->LoadModel<SourceModel>(*gbt_model));
+        return engine;
+      }
+
+      default:
+        return absl::InvalidArgumentError("Non supported GBDT model");
+    }
+  }
+};
+
+REGISTER_FastEngineFactory(
+    GradientBoostedTreesQuickScorerFastEngineFactory,
+    serving::gradient_boosted_trees::kQuickScorerExtended);
+
+class GradientBoostedTreesOptPredFastEngineFactory : public FastEngineFactory {
+ public:
+  using SourceModel = gradient_boosted_trees::GradientBoostedTreesModel;
+
+  std::string name() const override {
+    return serving::gradient_boosted_trees::kOptPred;
+  }
+
+  bool IsCompatible(const AbstractModel* const model) const override {
+    auto* gbt_model = dynamic_cast<const SourceModel*>(model);
+    if (gbt_model == nullptr) {
+      return false;
+    }
+
+    if (!gbt_model->IsMissingValueConditionResultFollowGlobalImputation()) {
+      return false;
+    }
+
+    for (const auto& src_tree : gbt_model->decision_trees()) {
+      if (src_tree->NumLeafs() > std::numeric_limits<uint16_t>::max()) {
+        return false;
+      }
+    }
+
+    for (const auto feature_idx : gbt_model->input_features()) {
+      const auto& feature = gbt_model->data_spec().columns(feature_idx);
+      switch (feature.type()) {
+        case dataset::proto::ColumnType::NUMERICAL:
+          break;
+        case dataset::proto::ColumnType::CATEGORICAL:
+          if (feature.categorical().number_of_unique_values() > 32) {
+            return false;
+          }
+          break;
+        default:
+          return false;
+      }
+    }
+
+    switch (gbt_model->task()) {
+      case proto::CLASSIFICATION:
+        return gbt_model->label_col_spec()
+                   .categorical()
+                   .number_of_unique_values() == 3;
+      case proto::REGRESSION:
+      case proto::RANKING:
+        return true;
+      default:
+        return false;
+    }
+  }
+
+  std::vector<std::string> IsBetterThan() const override {
+    return {serving::gradient_boosted_trees::kGeneric};
+  }
+
+  utils::StatusOr<std::unique_ptr<serving::FastEngine>> CreateEngine(
+      const AbstractModel* const model) const override {
+    auto* gbt_model = dynamic_cast<const SourceModel*>(model);
+    if (!gbt_model) {
+      return absl::InvalidArgumentError("The model is not a GBDT.");
+    }
+
+    switch (gbt_model->task()) {
+      case proto::CLASSIFICATION:
+        if (gbt_model->label_col_spec()
+                .categorical()
+                .number_of_unique_values() == 3) {
+          // Binary classification.
+          auto engine = absl::make_unique<serving::ExampleSetModelWrapper<
+              serving::decision_forest::
+                  GradientBoostedTreesBinaryClassificationNumericalAndCategorical,
+              serving::decision_forest::PredictWithExampleSet>>();
+          RETURN_IF_ERROR(engine->LoadModel<SourceModel>(*gbt_model));
+          return engine;
+        } else {
+          return absl::InvalidArgumentError("Non supported GBDT model");
+        }
+
+      case proto::REGRESSION: {
+        auto engine = absl::make_unique<serving::ExampleSetModelWrapper<
+            serving::decision_forest::
+                GradientBoostedTreesRegressionNumericalAndCategorical,
+            serving::decision_forest::PredictWithExampleSet>>();
+        RETURN_IF_ERROR(engine->LoadModel<SourceModel>(*gbt_model));
+        return engine;
+      }
+
+      case proto::RANKING: {
+        auto engine = absl::make_unique<serving::ExampleSetModelWrapper<
+            serving::decision_forest::
+                GradientBoostedTreesRankingNumericalAndCategorical,
+            serving::decision_forest::PredictWithExampleSet>>();
+        RETURN_IF_ERROR(engine->LoadModel<SourceModel>(*gbt_model));
+        return engine;
+      }
+
+      default:
+        return absl::InvalidArgumentError("Non supported GBDT model");
+    }
+  }
+};
+
+REGISTER_FastEngineFactory(GradientBoostedTreesOptPredFastEngineFactory,
+                           serving::gradient_boosted_trees::kOptPred);
+
 class RandomForestGenericFastEngineFactory : public model::FastEngineFactory {
  public:
   using SourceModel = random_forest::RandomForestModel;
@@ -222,5 +429,94 @@ class RandomForestGenericFastEngineFactory : public model::FastEngineFactory {
 REGISTER_FastEngineFactory(RandomForestGenericFastEngineFactory,
                            serving::random_forest::kGeneric);
 
+class RandomForestOptPredFastEngineFactory : public model::FastEngineFactory {
+ public:
+  using SourceModel = random_forest::RandomForestModel;
+
+  std::string name() const override { return serving::random_forest::kOptPred; }
+
+  bool IsCompatible(const AbstractModel* const model) const override {
+    auto* rf_model = dynamic_cast<const SourceModel*>(model);
+    // This implementation is the most generic and least efficient engine.
+    if (rf_model == nullptr) {
+      return false;
+    }
+    if (!rf_model->IsMissingValueConditionResultFollowGlobalImputation()) {
+      return false;
+    }
+
+    for (const auto& src_tree : rf_model->decision_trees()) {
+      if (src_tree->NumLeafs() > std::numeric_limits<uint16_t>::max()) {
+        return false;
+      }
+    }
+
+    for (const auto feature_idx : rf_model->input_features()) {
+      const auto& feature = rf_model->data_spec().columns(feature_idx);
+      switch (feature.type()) {
+        case dataset::proto::ColumnType::NUMERICAL:
+          break;
+        case dataset::proto::ColumnType::CATEGORICAL:
+          if (feature.categorical().number_of_unique_values() > 32) {
+            return false;
+          }
+          break;
+        default:
+          return false;
+      }
+    }
+
+    switch (rf_model->task()) {
+      case proto::CLASSIFICATION:
+        return rf_model->label_col_spec()
+                   .categorical()
+                   .number_of_unique_values() == 3;
+      case proto::REGRESSION:
+      case proto::RANKING:
+        return true;
+      default:
+        return false;
+    }
+  }
+
+  std::vector<std::string> IsBetterThan() const override {
+    return {serving::random_forest::kGeneric};
+  }
+
+  utils::StatusOr<std::unique_ptr<serving::FastEngine>> CreateEngine(
+      const AbstractModel* const model) const override {
+    auto* rf_model = dynamic_cast<const SourceModel*>(model);
+    if (!rf_model) {
+      return absl::InvalidArgumentError("The model is not a RF.");
+    }
+
+    switch (rf_model->task()) {
+      case model::proto::CLASSIFICATION: {
+        auto engine = absl::make_unique<serving::ExampleSetModelWrapper<
+            serving::decision_forest::
+                RandomForestBinaryClassificationNumericalAndCategoricalFeatures,
+            serving::decision_forest::PredictWithExampleSet>>();
+        RETURN_IF_ERROR(engine->LoadModel<SourceModel>(*rf_model));
+        return engine;
+      }
+
+      case model::proto::REGRESSION: {
+        auto engine = absl::make_unique<serving::ExampleSetModelWrapper<
+            serving::decision_forest::
+                RandomForestRegressionNumericalAndCategorical,
+            serving::decision_forest::PredictWithExampleSet>>();
+        RETURN_IF_ERROR(engine->LoadModel<SourceModel>(*rf_model));
+        return engine;
+      }
+
+      default:
+        return absl::InvalidArgumentError("Non supported RF model");
+    }
+  }
+};
+
+REGISTER_FastEngineFactory(RandomForestOptPredFastEngineFactory,
+                           serving::random_forest::kOptPred);
+
 }  // namespace model
 }  // namespace yggdrasil_decision_forests