Merge pull request #685 from flust/0.2.x

FEA: Add basic framework of 'lightgbm' and merge xgboost and lightgbm…

recbole/config/configurator.py

@@ -229,7 +229,7 @@ def _load_internal_config_dict(self, model, model_class, dataset):
         self.internal_config_dict['MODEL_TYPE'] = model_class.type
         if self.internal_config_dict['MODEL_TYPE'] == ModelType.GENERAL:
             pass
-        elif self.internal_config_dict['MODEL_TYPE'] in {ModelType.CONTEXT, ModelType.XGBOOST}:
+        elif self.internal_config_dict['MODEL_TYPE'] in {ModelType.CONTEXT, ModelType.DECISIONTREE}:
             self._update_internal_config_dict(context_aware_init)
             if dataset == 'ml-100k':
                 self._update_internal_config_dict(context_aware_on_ml_100k_init)

recbole/data/dataloader/__init__.py

@@ -4,5 +4,5 @@
 from recbole.data.dataloader.context_dataloader import *
 from recbole.data.dataloader.sequential_dataloader import *
 from recbole.data.dataloader.knowledge_dataloader import *
-from recbole.data.dataloader.xgboost_dataloader import *
+from recbole.data.dataloader.decisiontree_dataloader import *
 from recbole.data.dataloader.user_dataloader import *

recbole/data/dataloader/xgboost_dataloader.py → recbole/data/dataloader/decisiontree_dataloader.py

@@ -8,32 +8,32 @@
 # @Email  : [email protected]
 
 """
-recbole.data.dataloader.xgboost_dataloader
+recbole.data.dataloader.decisiontree_dataloader
 ################################################
 """
 
 from recbole.data.dataloader.general_dataloader import GeneralDataLoader, GeneralNegSampleDataLoader, \
     GeneralFullDataLoader
 
 
-class XgboostDataLoader(GeneralDataLoader):
-    """:class:`XgboostDataLoader` is inherit from
+class DecisionTreeDataLoader(GeneralDataLoader):
+    """:class:`DecisionTreeDataLoader` is inherit from
     :class:`~recbole.data.dataloader.general_dataloader.GeneralDataLoader`,
     and didn't add/change anything at all.
     """
     pass
 
 
-class XgboostNegSampleDataLoader(GeneralNegSampleDataLoader):
-    """:class:`XgboostNegSampleDataLoader` is inherit from
+class DecisionTreeNegSampleDataLoader(GeneralNegSampleDataLoader):
+    """:class:`DecisionTreeNegSampleDataLoader` is inherit from
     :class:`~recbole.data.dataloader.general_dataloader.GeneralNegSampleDataLoader`,
     and didn't add/change anything at all.
     """
     pass
 
 
-class XgboostFullDataLoader(GeneralFullDataLoader):
-    """:class:`XgboostFullDataLoader` is inherit from
+class DecisionTreeFullDataLoader(GeneralFullDataLoader):
+    """:class:`DecisionTreeFullDataLoader` is inherit from
     :class:`~recbole.data.dataloader.general_dataloader.GeneralFullDataLoader`,
     and didn't add/change anything at all.
     """

recbole/data/dataset/__init__.py

@@ -3,5 +3,5 @@
 from recbole.data.dataset.kg_dataset import KnowledgeBasedDataset
 from recbole.data.dataset.social_dataset import SocialDataset
 from recbole.data.dataset.kg_seq_dataset import Kg_Seq_Dataset
-from recbole.data.dataset.xgboost_dataset import XgboostDataset
+from recbole.data.dataset.decisiontree_dataset import DecisionTreeDataset
 from recbole.data.dataset.customized_dataset import *

recbole/data/dataset/xgboost_dataset.py → recbole/data/dataset/decisiontree_dataset.py

@@ -3,16 +3,16 @@
 # @Email  : [email protected]
 
 """
-recbole.data.xgboost_dataset
+recbole.data.decisiontree_dataset
 ##########################
 """
 
 from recbole.data.dataset import Dataset
 from recbole.utils import FeatureType
 
 
-class XgboostDataset(Dataset):
-    """:class:`XgboostDataset` is based on :class:`~recbole.data.dataset.dataset.Dataset`,
+class DecisionTreeDataset(Dataset):
+    """:class:`DecisionTreeDataset` is based on :class:`~recbole.data.dataset.dataset.Dataset`,
     and 
 
     Attributes:

recbole/data/utils.py

@@ -45,9 +45,9 @@ def create_dataset(config):
         elif model_type == ModelType.SOCIAL:
             from .dataset import SocialDataset
             return SocialDataset(config)
-        elif model_type == ModelType.XGBOOST:
-            from .dataset import XgboostDataset
-            return XgboostDataset(config)
+        elif model_type == ModelType.DECISIONTREE:
+            from .dataset import DecisionTreeDataset
+            return DecisionTreeDataset(config)
         else:
             from .dataset import Dataset
             return Dataset(config)
@@ -283,13 +283,13 @@ def get_data_loader(name, config, eval_setting):
             return SequentialNegSampleDataLoader
         elif neg_sample_strategy == 'full':
             return SequentialFullDataLoader
-    elif model_type == ModelType.XGBOOST:
+    elif model_type == ModelType.DECISIONTREE:
         if neg_sample_strategy == 'none':
-            return XgboostDataLoader
+            return DecisionTreeDataLoader
         elif neg_sample_strategy == 'by':
-            return XgboostNegSampleDataLoader
+            return DecisionTreeNegSampleDataLoader
         elif neg_sample_strategy == 'full':
-            return XgboostFullDataLoader
+            return DecisionTreeFullDataLoader
     elif model_type == ModelType.KNOWLEDGE:
         if neg_sample_strategy == 'by':
             if name == 'train':

recbole/model/exlib_recommender/lightgbm.py

@@ -0,0 +1,26 @@
+# -*- coding: utf-8 -*-
+# @Time   : 2020/1/17
+# @Author : Chen Yang
+# @Email  : [email protected]
+
+r"""
+recbole.model.exlib_recommender.lightgbm
+#############################
+"""
+
+import lightgbm as lgb
+from recbole.utils import ModelType, InputType
+
+
+class lightgbm(lgb.Booster):
+    r"""lightgbm is inherited from lgb.Booster
+
+    """
+    type = ModelType.DECISIONTREE
+    input_type = InputType.POINTWISE
+
+    def __init__(self, config, dataset):
+        super(lgb.Booster, self).__init__()
+
+    def to(self, device):
+        return self

recbole/model/exlib_recommender/xgboost.py

@@ -16,7 +16,7 @@ class xgboost(xgb.Booster):
     r"""xgboost is inherited from xgb.Booster
 
     """
-    type = ModelType.XGBOOST
+    type = ModelType.DECISIONTREE
     input_type = InputType.POINTWISE
 
     def __init__(self, config, dataset):

recbole/properties/model/lightgbm.yaml

@@ -0,0 +1,23 @@
+convert_token_to_onehot: False
+token_num_threshold: 10000
+
+# Dataset
+lgb_silent: False
+
+# Train
+lgb_model: ~
+lgb_params: 
+    boosting: gbdt
+    num_leaves: 90
+    min_data_in_leaf: 30
+    max_depth: -1
+    learning_rate: 0.1
+    objective: binary
+    lambda_l1: 0.1
+    metric: ['auc', 'binary_logloss']
+    force_row_wise: True
+lgb_learning_rates: ~
+lgb_num_boost_round: 300
+lgb_early_stopping_rounds: ~
+lgb_verbose_eval: 100
+

recbole/properties/model/xgboost.yaml

@@ -1,14 +1,8 @@
-# Type of training method
 convert_token_to_onehot: False
 token_num_threshold: 10000
 
 # DMatrix
-xgb_weight: ~
-xgb_base_margin: ~
-xgb_missing: ~
 xgb_silent: ~
-xgb_feature_names: ~
-xgb_feature_types: ~
 xgb_nthread: ~
 
 xgb_model: ~
@@ -26,11 +20,6 @@ xgb_params:
     seed: 2020
     # nthread: -1
 xgb_num_boost_round: 500
-# xgb_evals: ~
-xgb_obj: ~
-xgb_feval: ~
-xgb_maximize: ~
 xgb_early_stopping_rounds: ~
-# xgb_evals_result: ~
 xgb_verbose_eval: 100