lightgbm 0.1.3 → 0.1.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 3d841acf71e8af7111178da8c2062b47900ec953a94154a0cdf9f28bf7d61714
-  data.tar.gz: 6ed019f4094803a06be77008e48870fb8db3acac4b83f3675eaeae4e20c27fdb
+  metadata.gz: 81f1f695112234bb576afaab35f4bf276d1f9c4a4adf0c74831cd1bb73f6baa0
+  data.tar.gz: 59ef1f3c581f83e108ce2a6f2c847bb7488fc2ea7f39ba217aaeffbf46e99351
 SHA512:
-  metadata.gz: 477e25066789028e7b8a8a78107c1ed823bd06d96d97afdda41b502e2e3e4a9e0065888c414effe4ace4097baa4d4b18988c4ee6b4a9d06347992afa201a52b5
-  data.tar.gz: eabb924994ffcafce6cb9038a60e3327528d2308d39c62bc336a06191e471ff412e141f9117446abc068aabbe9d1d16be59cc8bdca889270219895e85ec9e57b
+  metadata.gz: 21297d26e88957dd60d0aa61da19aa53aa632958adaf069b0efe2c8dae35e2e21e74c374da3509e337ca3268613b14dc541aee5012df086af7e8f784adb5063d
+  data.tar.gz: 7dbdc0fccaf256a1a835aea3eaa51fe326a0cd4b8cde168a6b8c27ff00c6412a5b5d82583fb1c14749f001f002a4fa3c3e156b9c4b2174b1e91f9449a4fa9ba1

data/CHANGELOG.md

@@ -1,3 +1,11 @@
+## 0.1.4
+
+- Friendlier message when LightGBM not found
+- Added `Ranker`
+- Added early stopping to Scikit-Learn API
+- Free memory when objects are destroyed
+- Removed unreleased `dump_text` method
+
 ## 0.1.3
 
 - Added Scikit-Learn API

data/README.md

@@ -18,7 +18,7 @@ gem 'lightgbm'
 
 ## Getting Started
 
-This library follows the [Data Structure, Training, and Scikit-Learn APIs](https://lightgbm.readthedocs.io/en/latest/Python-API.html) of the Python library. A few differences are:
+This library follows the [Python API](https://lightgbm.readthedocs.io/en/latest/Python-API.html). A few differences are:
 
 - The `get_` prefix is removed from methods
 - The default verbosity is `-1`
@@ -63,7 +63,7 @@ booster.feature_importance
 Early stopping
 
 ```ruby
-LightGBM.train(params, train_set, valid_set: [train_set, test_set], early_stopping_rounds: 5)
+LightGBM.train(params, train_set, valid_sets: [train_set, test_set], early_stopping_rounds: 5)
 ```
 
 CV
@@ -116,6 +116,12 @@ Get the importance of features
 model.feature_importances
 ```
 
+Early stopping
+
+```ruby
+model.fit(x, y, eval_set: [[x_test, y_test]], early_stopping_rounds: 5)
+```
+
 ## Data
 
 Data can be an array of arrays

data/lib/lightgbm.rb

@@ -5,16 +5,26 @@ require "ffi"
 require "lightgbm/utils"
 require "lightgbm/booster"
 require "lightgbm/dataset"
-require "lightgbm/ffi"
 require "lightgbm/version"
 
 # scikit-learn API
+require "lightgbm/model"
 require "lightgbm/classifier"
+require "lightgbm/ranker"
 require "lightgbm/regressor"
 
 module LightGBM
   class Error < StandardError; end
 
+  class << self
+    attr_accessor :ffi_lib
+  end
+  lib_name = "lib_lightgbm.#{::FFI::Platform::LIBSUFFIX}"
+  self.ffi_lib = [lib_name, "lib_lightgbm.so"]
+
+  # friendlier error message
+  autoload :FFI, "lightgbm/ffi"
+
   class << self
     def train(params, train_set, num_boost_round: 100, valid_sets: [], valid_names: [], early_stopping_rounds: nil, verbose_eval: true)
       booster = Booster.new(params: params, train_set: train_set)

data/lib/lightgbm/booster.rb

@@ -14,8 +14,7 @@ module LightGBM
         set_verbosity(params)
         check_result FFI.LGBM_BoosterCreate(train_set.handle_pointer, params_str(params), @handle)
       end
-      # causes "Stack consistency error"
-      # ObjectSpace.define_finalizer(self, self.class.finalize(handle_pointer))
+      ObjectSpace.define_finalizer(self, self.class.finalize(handle_pointer))
 
       self.best_iteration = -1
 
@@ -23,10 +22,6 @@ module LightGBM
       @name_valid_sets = []
     end
 
-    def self.finalize(pointer)
-      -> { FFI.LGBM_BoosterFree(pointer) }
-    end
-
     def add_valid(data, name)
       check_result FFI.LGBM_BoosterAddValidData(handle_pointer, data.handle_pointer)
       @name_valid_sets << name
@@ -153,6 +148,11 @@ module LightGBM
       finished.read_int == 1
     end
 
+    def self.finalize(pointer)
+      # must use proc instead of stabby lambda
+      proc { FFI.LGBM_BoosterFree(pointer) }
+    end
+
     private
 
     def handle_pointer

data/lib/lightgbm/classifier.rb

@@ -1,15 +1,10 @@
 module LightGBM
-  class Classifier
-    def initialize(num_leaves: 31, learning_rate: 0.1, n_estimators: 100, objective: nil)
-      @params = {
-        num_leaves: num_leaves,
-        learning_rate: learning_rate
-      }
-      @params[:objective] = objective if objective
-      @n_estimators = n_estimators
+  class Classifier < Model
+    def initialize(num_leaves: 31, learning_rate: 0.1, n_estimators: 100, objective: nil, **options)
+      super
     end
 
-    def fit(x, y)
+    def fit(x, y, eval_set: nil, eval_names: [], categorical_feature: "auto", early_stopping_rounds: nil, verbose: true)
       n_classes = y.uniq.size
 
       params = @params.dup
@@ -20,26 +15,34 @@ module LightGBM
         params[:objective] ||= "binary"
       end
 
-      train_set = Dataset.new(x, label: y)
-      @booster = LightGBM.train(params, train_set, num_boost_round: @n_estimators)
+      train_set = Dataset.new(x, label: y, categorical_feature: categorical_feature)
+      valid_sets = Array(eval_set).map { |v| Dataset.new(v[0], label: v[1], reference: train_set) }
+
+      @booster = LightGBM.train(params, train_set,
+        num_boost_round: @n_estimators,
+        early_stopping_rounds: early_stopping_rounds,
+        verbose_eval: verbose,
+        valid_sets: valid_sets,
+        valid_names: eval_names
+      )
       nil
     end
 
-    def predict(data)
-      y_pred = @booster.predict(data)
+    def predict(data, num_iteration: nil)
+      y_pred = @booster.predict(data, num_iteration: num_iteration)
 
       if y_pred.first.is_a?(Array)
         # multiple classes
         y_pred.map do |v|
-          v.map.with_index.max_by { |v2, i| v2 }.last
+          v.map.with_index.max_by { |v2, _| v2 }.last
         end
       else
         y_pred.map { |v| v > 0.5 ? 1 : 0 }
       end
     end
 
-    def predict_proba(data)
-      y_pred = @booster.predict(data)
+    def predict_proba(data, num_iteration: nil)
+      y_pred = @booster.predict(data, num_iteration: num_iteration)
 
       if y_pred.first.is_a?(Array)
         # multiple classes
@@ -48,17 +51,5 @@ module LightGBM
         y_pred.map { |v| [1 - v, v] }
       end
     end
-
-    def save_model(fname)
-      @booster.save_model(fname)
-    end
-
-    def load_model(fname)
-      @booster = Booster.new(params: @params, model_file: fname)
-    end
-
-    def feature_importances
-      @booster.feature_importance
-    end
   end
 end

data/lib/lightgbm/dataset.rb

@@ -2,7 +2,7 @@ module LightGBM
   class Dataset
     attr_reader :data, :params
 
-    def initialize(data, label: nil, weight: nil, params: nil, reference: nil, used_indices: nil, categorical_feature: "auto")
+    def initialize(data, label: nil, weight: nil, group: nil, params: nil, reference: nil, used_indices: nil, categorical_feature: "auto")
       @data = data
 
       # TODO stringify params
@@ -13,12 +13,12 @@ module LightGBM
       @handle = ::FFI::MemoryPointer.new(:pointer)
       parameters = params_str(params)
       reference = reference.handle_pointer if reference
-      if data.is_a?(String)
-        check_result FFI.LGBM_DatasetCreateFromFile(data, parameters, reference, @handle)
-      elsif used_indices
+      if used_indices
         used_row_indices = ::FFI::MemoryPointer.new(:int32, used_indices.count)
         used_row_indices.put_array_of_int32(0, used_indices)
         check_result FFI.LGBM_DatasetGetSubset(reference, used_row_indices, used_indices.count, parameters, @handle)
+      elsif data.is_a?(String)
+        check_result FFI.LGBM_DatasetCreateFromFile(data, parameters, reference, @handle)
       else
         if matrix?(data)
           nrow = data.row_count
@@ -40,11 +40,11 @@ module LightGBM
         c_data.put_array_of_float(0, flat_data)
         check_result FFI.LGBM_DatasetCreateFromMat(c_data, 0, nrow, ncol, 1, parameters, reference, @handle)
       end
-      # causes "Stack consistency error"
-      # ObjectSpace.define_finalizer(self, self.class.finalize(handle_pointer))
+      ObjectSpace.define_finalizer(self, self.class.finalize(handle_pointer)) unless used_indices
 
-      set_field("label", label) if label
-      set_field("weight", weight) if weight
+      self.label = label if label
+      self.weight = weight if weight
+      self.group = group if group
     end
 
     def label
@@ -55,6 +55,18 @@ module LightGBM
       field("weight")
     end
 
+    def label=(label)
+      set_field("label", label)
+    end
+
+    def weight=(weight)
+      set_field("weight", weight)
+    end
+
+    def group=(group)
+      set_field("group", group, type: :int32)
+    end
+
     def num_data
       out = ::FFI::MemoryPointer.new(:int)
       check_result FFI.LGBM_DatasetGetNumData(handle_pointer, out)
@@ -71,9 +83,10 @@ module LightGBM
       check_result FFI.LGBM_DatasetSaveBinary(handle_pointer, filename)
     end
 
-    def dump_text(filename)
-      check_result FFI.LGBM_DatasetDumpText(handle_pointer, filename)
-    end
+    # not released yet
+    # def dump_text(filename)
+    #   check_result FFI.LGBM_DatasetDumpText(handle_pointer, filename)
+    # end
 
     def subset(used_indices, params: nil)
       # categorical_feature passed via params
@@ -85,14 +98,15 @@ module LightGBM
       )
     end
 
-    def self.finalize(pointer)
-      -> { FFI.LGBM_DatasetFree(pointer) }
-    end
-
     def handle_pointer
       @handle.read_pointer
     end
 
+    def self.finalize(pointer)
+      # must use proc instead of stabby lambda
+      proc { FFI.LGBM_DatasetFree(pointer) }
+    end
+
     private
 
     def field(field_name)
@@ -104,11 +118,17 @@ module LightGBM
       out_ptr.read_pointer.read_array_of_float(num_data)
     end
 
-    def set_field(field_name, data)
+    def set_field(field_name, data, type: :float)
       data = data.to_a unless data.is_a?(Array)
-      c_data = ::FFI::MemoryPointer.new(:float, data.count)
-      c_data.put_array_of_float(0, data)
-      check_result FFI.LGBM_DatasetSetField(handle_pointer, field_name, c_data, data.count, 0)
+      if type == :int32
+        c_data = ::FFI::MemoryPointer.new(:int32, data.count)
+        c_data.put_array_of_int32(0, data)
+        check_result FFI.LGBM_DatasetSetField(handle_pointer, field_name, c_data, data.count, 2)
+      else
+        c_data = ::FFI::MemoryPointer.new(:float, data.count)
+        c_data.put_array_of_float(0, data)
+        check_result FFI.LGBM_DatasetSetField(handle_pointer, field_name, c_data, data.count, 0)
+      end
     end
 
     def matrix?(data)

data/lib/lightgbm/ffi.rb

@@ -1,7 +1,13 @@
 module LightGBM
   module FFI
     extend ::FFI::Library
-    ffi_lib ["lightgbm", "lib_lightgbm.so"]
+
+    begin
+      ffi_lib LightGBM.ffi_lib
+    rescue LoadError => e
+      raise e if ENV["LIGHTGBM_DEBUG"]
+      raise LoadError, "Could not find LightGBM"
+    end
 
     # https://github.com/microsoft/LightGBM/blob/master/include/LightGBM/c_api.h
     # keep same order
@@ -15,7 +21,7 @@ module LightGBM
     attach_function :LGBM_DatasetGetSubset, %i[pointer pointer int32 string pointer], :int
     attach_function :LGBM_DatasetFree, %i[pointer], :int
     attach_function :LGBM_DatasetSaveBinary, %i[pointer string], :int
-    attach_function :LGBM_DatasetDumpText, %i[pointer string], :int
+    # attach_function :LGBM_DatasetDumpText, %i[pointer string], :int
     attach_function :LGBM_DatasetSetField, %i[pointer string pointer int int], :int
     attach_function :LGBM_DatasetGetField, %i[pointer string pointer pointer pointer], :int
     attach_function :LGBM_DatasetGetNumData, %i[pointer pointer], :int

data/lib/lightgbm/model.rb

@@ -0,0 +1,30 @@
+module LightGBM
+  class Model
+    attr_reader :booster
+
+    def initialize(num_leaves: 31, learning_rate: 0.1, n_estimators: 100, objective: nil, **options)
+      @params = {
+        num_leaves: num_leaves,
+        learning_rate: learning_rate
+      }.merge(options)
+      @params[:objective] = objective if objective
+      @n_estimators = n_estimators
+    end
+
+    def save_model(fname)
+      @booster.save_model(fname)
+    end
+
+    def load_model(fname)
+      @booster = Booster.new(params: @params, model_file: fname)
+    end
+
+    def best_iteration
+      @booster.best_iteration
+    end
+
+    def feature_importances
+      @booster.feature_importance
+    end
+  end
+end

data/lib/lightgbm/ranker.rb

@@ -0,0 +1,21 @@
+module LightGBM
+  class Ranker < Model
+    def initialize(num_leaves: 31, learning_rate: 0.1, n_estimators: 100, objective: "lambdarank", **options)
+      super
+    end
+
+    def fit(x, y, group:, categorical_feature: "auto", early_stopping_rounds: nil, verbose: true)
+      train_set = Dataset.new(x, label: y, group: group, categorical_feature: categorical_feature)
+      @booster = LightGBM.train(@params, train_set,
+        num_boost_round: @n_estimators,
+        early_stopping_rounds: early_stopping_rounds,
+        verbose_eval: verbose
+      )
+      nil
+    end
+
+    def predict(data, num_iteration: nil)
+      @booster.predict(data, num_iteration: num_iteration)
+    end
+  end
+end

data/lib/lightgbm/regressor.rb

@@ -1,34 +1,25 @@
 module LightGBM
-  class Regressor
-    def initialize(num_leaves: 31, learning_rate: 0.1, n_estimators: 100, objective: nil)
-      @params = {
-        num_leaves: num_leaves,
-        learning_rate: learning_rate
-      }
-      @params[:objective] = objective if objective
-      @n_estimators = n_estimators
+  class Regressor < Model
+    def initialize(num_leaves: 31, learning_rate: 0.1, n_estimators: 100, objective: "regression", **options)
+      super
     end
 
-    def fit(x, y)
-      train_set = Dataset.new(x, label: y)
-      @booster = LightGBM.train(@params, train_set, num_boost_round: @n_estimators)
-      nil
-    end
-
-    def predict(data)
-      @booster.predict(data)
-    end
+    def fit(x, y, categorical_feature: "auto", eval_set: nil, eval_names: [], early_stopping_rounds: nil, verbose: true)
+      train_set = Dataset.new(x, label: y, categorical_feature: categorical_feature)
+      valid_sets = Array(eval_set).map { |v| Dataset.new(v[0], label: v[1], reference: train_set) }
 
-    def save_model(fname)
-      @booster.save_model(fname)
-    end
-
-    def load_model(fname)
-      @booster = Booster.new(params: @params, model_file: fname)
+      @booster = LightGBM.train(@params, train_set,
+        num_boost_round: @n_estimators,
+        early_stopping_rounds: early_stopping_rounds,
+        verbose_eval: verbose,
+        valid_sets: valid_sets,
+        valid_names: eval_names
+      )
+      nil
     end
 
-    def feature_importances
-      @booster.feature_importance
+    def predict(data, num_iteration: nil)
+      @booster.predict(data, num_iteration: num_iteration)
     end
   end
 end

data/lib/lightgbm/version.rb

@@ -1,3 +1,3 @@
 module LightGBM
-  VERSION = "0.1.3"
+  VERSION = "0.1.4"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: lightgbm
 version: !ruby/object:Gem::Version
-  version: 0.1.3
+  version: 0.1.4
 platform: ruby
 authors:
 - Andrew Kane
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2019-08-16 00:00:00.000000000 Z
+date: 2019-08-19 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: ffi
@@ -107,6 +107,8 @@ files:
 - lib/lightgbm/classifier.rb
 - lib/lightgbm/dataset.rb
 - lib/lightgbm/ffi.rb
+- lib/lightgbm/model.rb
+- lib/lightgbm/ranker.rb
 - lib/lightgbm/regressor.rb
 - lib/lightgbm/utils.rb
 - lib/lightgbm/version.rb
@@ -129,7 +131,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.0.4
+rubygems_version: 3.0.3
 signing_key: 
 specification_version: 4
 summary: LightGBM - the high performance machine learning library - for Ruby