lightgbm 0.1.2 → 0.1.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 723130d41ea9196bcbd7bcffeb865c40c65985f26eca018d49bd176d33c43142
-  data.tar.gz: d92b41899ff72da2ef4e5782bf4d2840caee1554107d9fd5d02bd6728829585a
+  metadata.gz: 3d841acf71e8af7111178da8c2062b47900ec953a94154a0cdf9f28bf7d61714
+  data.tar.gz: 6ed019f4094803a06be77008e48870fb8db3acac4b83f3675eaeae4e20c27fdb
 SHA512:
-  metadata.gz: 6960dbf1e2a884705e8a2752952392483c0ab1e74e970382b45df747ccbedbc0d978d3910e2b935190c98a9c45315841b53e27128aa5944e3a7834808e05582a
-  data.tar.gz: c6e793933dc794fa62099580ad35f29d7e5e3ae24a07df4335dca3d68571bc1d5b360a7cf0859d75d245a77d470a0dafba0cb5d86edda1974f3bc532b0f5c11a
+  metadata.gz: 477e25066789028e7b8a8a78107c1ed823bd06d96d97afdda41b502e2e3e4a9e0065888c414effe4ace4097baa4d4b18988c4ee6b4a9d06347992afa201a52b5
+  data.tar.gz: eabb924994ffcafce6cb9038a60e3327528d2308d39c62bc336a06191e471ff412e141f9117446abc068aabbe9d1d16be59cc8bdca889270219895e85ec9e57b

data/CHANGELOG.md

@@ -1,3 +1,8 @@
+## 0.1.3
+
+- Added Scikit-Learn API
+- Added support for Daru and Numo::NArray
+
 ## 0.1.2
 
 - Added `cv` method

data/README.md

@@ -1,6 +1,6 @@
 # LightGBM
 
-[LightGBM](https://github.com/microsoft/LightGBM) for Ruby
+[LightGBM](https://github.com/microsoft/LightGBM) - the high performance machine learning library - for Ruby
 
 :fire: Uses the C API for blazing performance
 
@@ -18,6 +18,16 @@ 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:
+
+- The `get_` prefix is removed from methods
+- The default verbosity is `-1`
+- With the `cv` method, `stratified` is set to `false`
+
+Some methods and options are also missing at the moment. PRs welcome!
+
+## Training API
+
 Train a model
 
 ```ruby
@@ -44,38 +54,98 @@ Load the model from a file
 booster = LightGBM::Booster.new(model_file: "model.txt")
 ```
 
-Get feature importance
+Get the importance of features
 
 ```ruby
 booster.feature_importance
 ```
 
-## Early Stopping
+Early stopping
 
 ```ruby
 LightGBM.train(params, train_set, valid_set: [train_set, test_set], early_stopping_rounds: 5)
 ```
 
-## CV
+CV
 
 ```ruby
 LightGBM.cv(params, train_set, nfold: 5, verbose_eval: true)
 ```
 
-## Reference
+## Scikit-Learn API
 
-This library follows the [Data Structure and Training APIs](https://lightgbm.readthedocs.io/en/latest/Python-API.html) for the Python library. A few differences are:
+Prep your data
 
-- The default verbosity is `-1`
-- With the `cv` method, `stratified` is set to `false`
+```ruby
+x = [[1, 2], [3, 4], [5, 6], [7, 8]]
+y = [1, 2, 3, 4]
+```
 
-Some methods and options are also missing at the moment. PRs welcome!
+Train a model
+
+```ruby
+model = LightGBM::Regressor.new
+model.fit(x, y)
+```
+
+> For classification, use `LightGBM::Classifier`
+
+Predict
+
+```ruby
+model.predict(x)
+```
+
+> For classification, use `predict_proba` for probabilities
+
+Save the model to a file
+
+```ruby
+model.save_model("model.txt")
+```
+
+Load the model from a file
+
+```ruby
+model.load_model("model.txt")
+```
+
+Get the importance of features
+
+```ruby
+model.feature_importances
+```
+
+## Data
+
+Data can be an array of arrays
+
+```ruby
+[[1, 2, 3], [4, 5, 6]]
+```
+
+Or a Daru data frame
+
+```ruby
+Daru::DataFrame.from_csv("houses.csv")
+```
+
+Or a Numo NArray
+
+```ruby
+Numo::DFloat.new(3, 2).seq
+```
 
 ## Helpful Resources
 
 - [Parameters](https://lightgbm.readthedocs.io/en/latest/Parameters.html)
 - [Parameter Tuning](https://lightgbm.readthedocs.io/en/latest/Parameters-Tuning.html)
 
+## Related Projects
+
+- [Xgb](https://github.com/ankane/xgb) - XGBoost for Ruby
+- [Eps](https://github.com/ankane/eps) - Machine Learning for Ruby
+
 ## Credits
 
 Thanks to the [xgboost](https://github.com/PairOnAir/xgboost-ruby) gem for serving as an initial reference, and Selva Prabhakaran for the [test datasets](https://github.com/selva86/datasets).

data/lib/lightgbm.rb

@@ -8,11 +8,15 @@ require "lightgbm/dataset"
 require "lightgbm/ffi"
 require "lightgbm/version"
 
+# scikit-learn API
+require "lightgbm/classifier"
+require "lightgbm/regressor"
+
 module LightGBM
   class Error < StandardError; end
 
   class << self
-    def train(params, train_set,num_boost_round: 100, valid_sets: [], valid_names: [], early_stopping_rounds: nil, verbose_eval: true)
+    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)
 
       valid_contain_train = false
@@ -150,6 +154,7 @@ module LightGBM
         if early_stopping_rounds
           stop_early = false
           means.each do |k, score|
+            # TODO fix higher better
             if best_score[k].nil? || score < best_score[k]
               best_score[k] = score
               best_iter[k] = iteration

data/lib/lightgbm/booster.rb

@@ -77,7 +77,7 @@ module LightGBM
       num_feature = self.num_feature
       out_result = ::FFI::MemoryPointer.new(:double, num_feature)
       check_result FFI.LGBM_BoosterFeatureImportance(handle_pointer, iteration, importance_type, out_result)
-      out_result.read_array_of_double(num_feature)
+      out_result.read_array_of_double(num_feature).map(&:to_i)
     end
 
     def model_from_string(model_str)

data/lib/lightgbm/classifier.rb

@@ -0,0 +1,64 @@
+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
+    end
+
+    def fit(x, y)
+      n_classes = y.uniq.size
+
+      params = @params.dup
+      if n_classes > 2
+        params[:objective] ||= "multiclass"
+        params[:num_class] = n_classes
+      else
+        params[:objective] ||= "binary"
+      end
+
+      train_set = Dataset.new(x, label: y)
+      @booster = LightGBM.train(params, train_set, num_boost_round: @n_estimators)
+      nil
+    end
+
+    def predict(data)
+      y_pred = @booster.predict(data)
+
+      if y_pred.first.is_a?(Array)
+        # multiple classes
+        y_pred.map do |v|
+          v.map.with_index.max_by { |v2, i| v2 }.last
+        end
+      else
+        y_pred.map { |v| v > 0.5 ? 1 : 0 }
+      end
+    end
+
+    def predict_proba(data)
+      y_pred = @booster.predict(data)
+
+      if y_pred.first.is_a?(Array)
+        # multiple classes
+        y_pred
+      else
+        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

@@ -20,9 +20,25 @@ module LightGBM
         used_row_indices.put_array_of_int32(0, used_indices)
         check_result FFI.LGBM_DatasetGetSubset(reference, used_row_indices, used_indices.count, parameters, @handle)
       else
-        c_data = ::FFI::MemoryPointer.new(:float, data.count * data.first.count)
-        c_data.put_array_of_float(0, data.flatten)
-        check_result FFI.LGBM_DatasetCreateFromMat(c_data, 0, data.count, data.first.count, 1, parameters, reference, @handle)
+        if matrix?(data)
+          nrow = data.row_count
+          ncol = data.column_count
+          flat_data = data.to_a.flatten
+        elsif daru?(data)
+          nrow, ncol = data.shape
+          flat_data = data.each_vector.map(&:to_a).flatten
+        elsif narray?(data)
+          nrow, ncol = data.shape
+          flat_data = data.flatten.to_a
+        else
+          nrow = data.count
+          ncol = data.first.count
+          flat_data = data.flatten
+        end
+
+        c_data = ::FFI::MemoryPointer.new(:float, nrow * ncol)
+        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))
@@ -89,11 +105,24 @@ module LightGBM
     end
 
     def set_field(field_name, data)
+      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)
     end
 
+    def matrix?(data)
+      defined?(Matrix) && data.is_a?(Matrix)
+    end
+
+    def daru?(data)
+      defined?(Daru::DataFrame) && data.is_a?(Daru::DataFrame)
+    end
+
+    def narray?(data)
+      defined?(Numo::NArray) && data.is_a?(Numo::NArray)
+    end
+
     include Utils
   end
 end

data/lib/lightgbm/regressor.rb

@@ -0,0 +1,34 @@
+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
+    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 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/version.rb

@@ -1,3 +1,3 @@
 module LightGBM
-  VERSION = "0.1.2"
+  VERSION = "0.1.3"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: lightgbm
 version: !ruby/object:Gem::Version
-  version: 0.1.2
+  version: 0.1.3
 platform: ruby
 authors:
 - Andrew Kane
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2019-08-15 00:00:00.000000000 Z
+date: 2019-08-16 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: ffi
@@ -66,6 +66,34 @@ dependencies:
     - - ">="
       - !ruby/object:Gem::Version
         version: '5'
+- !ruby/object:Gem::Dependency
+  name: daru
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+- !ruby/object:Gem::Dependency
+  name: numo-narray
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
 description: 
 email: andrew@chartkick.com
 executables: []
@@ -76,8 +104,10 @@ files:
 - README.md
 - lib/lightgbm.rb
 - lib/lightgbm/booster.rb
+- lib/lightgbm/classifier.rb
 - lib/lightgbm/dataset.rb
 - lib/lightgbm/ffi.rb
+- lib/lightgbm/regressor.rb
 - lib/lightgbm/utils.rb
 - lib/lightgbm/version.rb
 homepage: https://github.com/ankane/lightgbm
@@ -102,5 +132,5 @@ requirements: []
 rubygems_version: 3.0.4
 signing_key: 
 specification_version: 4
-summary: LightGBM for Ruby
+summary: LightGBM - the high performance machine learning library - for Ruby
 test_files: []