Merge pull request EducationalTestingService#118 from nineil/feature/…

…add_sgd Add SGD
nineil · Jun 12, 2014 · 3f25150 · 3f25150
2 parents 29a3a3a + 7bbe8da
commit 3f25150
Show file tree

Hide file tree

Showing 3 changed files with 28 additions and 10 deletions.
diff --git a/skll/learner.py b/skll/learner.py
@@ -40,11 +40,11 @@
                               GradientBoostingRegressor, RandomForestClassifier,
                               RandomForestRegressor)
 from sklearn.linear_model import (ElasticNet, Lasso, LinearRegression,
-                                  LogisticRegression, Ridge)
+                                  LogisticRegression, Ridge, SGDClassifier,
+                                  SGDRegressor)
 from sklearn.naive_bayes import MultinomialNB
 from sklearn.svm import LinearSVC, SVC, SVR
 from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
-
 from skll.data import ExamplesTuple
 from skll.metrics import _CORRELATION_METRICS, use_score_func
 from skll.version import VERSION
@@ -74,15 +74,26 @@
                         'ElasticNet': [{'alpha': [0.01, 0.1, 1.0, 10.0,
                                                   100.0]}],
                         'SVR': [{'C': [0.01, 0.1, 1.0, 10.0, 100.0]}],
-                        'LinearRegression': [{}]}
+                        'LinearRegression': [{}],
+                        'SGDClassifier': [{'alpha': [0.000001, 0.00001, 0.0001,
+                                                     0.001, 0.01],
+                                           'penalty': ['l1', 'l2',
+                                                       'elasticnet']}],
+                        'SGDRegressor': [{'alpha': [0.000001, 0.00001, 0.0001,
+                                                    0.001, 0.01],
+                                          'penalty': ['l1', 'l2',
+                                                      'elasticnet']}]}
+
 _REGRESSION_MODELS = frozenset(['DecisionTreeRegressor', 'ElasticNet',
                                 'GradientBoostingRegressor', 'Lasso',
                                 'LinearRegression', 'RandomForestRegressor',
-                                'Ridge', 'SVR'])
+                                'Ridge', 'SVR', 'SGDRegressor'])
+
 _REQUIRES_DENSE = frozenset(['DecisionTreeClassifier', 'DecisionTreeRegressor',
                              'GradientBoostingClassifier',
                              'GradientBoostingRegressor', 'MultinomialNB',
                              'RandomForestClassifier', 'RandomForestRegressor'])
+
 MAX_CONCURRENT_PROCESSES = int(os.getenv('SKLL_MAX_CONCURRENT_PROCESSES', '5'))
 
 
@@ -301,6 +312,7 @@ def init(self, constrain=True, rescale=True, **kwargs):
     # Return modified class
     return cls
 
+
 # Rescaled regressors
 @rescaled
 class RescaledDecisionTreeRegressor(DecisionTreeRegressor):
@@ -342,6 +354,11 @@ class RescaledSVR(SVR):
     pass
 
 
+@rescaled
+class RescaledSGDRegressor(SGDRegressor):
+    pass
+
+
 class Learner(object):
     """
     A simpler learner interface around many scikit-learn classification
@@ -427,13 +444,15 @@ def __init__(self, model_type, probability=False, feature_scaling='none',
         elif self._model_type == 'SVR':
             self._model_kwargs['cache_size'] = 1000
             self._model_kwargs['kernel'] = 'linear'
-
+        elif self._model_type == 'SGDClassifier':
+            self._model_kwargs['loss'] = 'log'
         if self._model_type in {'RandomForestClassifier', 'LinearSVC',
                                 'LogisticRegression', 'DecisionTreeClassifier',
                                 'GradientBoostingClassifier',
                                 'GradientBoostingRegressor',
                                 'DecisionTreeRegressor',
-                                'RandomForestRegressor'}:
+                                'RandomForestRegressor', 'SGDClassifier',
+                                'SGDRegressor'}:
             self._model_kwargs['random_state'] = 123456789
 
         if model_kwargs:
@@ -523,7 +542,7 @@ def model_params(self):
             for feat, idx in iteritems(self.feat_vectorizer.vocabulary_):
                 if coef[idx]:
                     res[feat] = correction * coef[idx]
-                    #res[feat] = coef[idx]
+                    # res[feat] = coef[idx]
         elif isinstance(self._model, BaseLibLinear):
             label_list = self.label_list
 
@@ -677,7 +696,6 @@ def _train_setup(self, examples):
                                              with_mean=False,
                                              with_std=False)
 
-
     def train(self, examples, param_grid=None, grid_search_folds=5,
               grid_search=True, grid_objective='f1_score_micro',
               grid_jobs=None, shuffle=True):

diff --git a/tests/configs/test_regression1.template.cfg b/tests/configs/test_regression1.template.cfg
@@ -5,7 +5,7 @@ task=evaluate
 [Input]
 train_location=
 featuresets=[["test_regression1"]]
-learners = ['DecisionTreeRegressor', 'GradientBoostingRegressor', 'RandomForestRegressor', 'Ridge', 'SVR', 'RescaledDecisionTreeRegressor', 'RescaledGradientBoostingRegressor', 'RescaledRandomForestRegressor', 'RescaledRidge', 'RescaledSVR', 'Lasso', 'RescaledLasso', 'LinearRegression', 'RescaledLinearRegression', 'ElasticNet', 'RescaledElasticNet']
+learners = ['DecisionTreeRegressor', 'GradientBoostingRegressor', 'RandomForestRegressor', 'Ridge', 'SVR', 'RescaledDecisionTreeRegressor', 'RescaledGradientBoostingRegressor', 'RescaledRandomForestRegressor', 'RescaledRidge', 'RescaledSVR', 'Lasso', 'RescaledLasso', 'LinearRegression', 'RescaledLinearRegression', 'ElasticNet', 'RescaledElasticNet', 'SGDRegressor', 'RescaledSGDRegressor']
 suffix=.jsonlines
 
 [Tuning]

diff --git a/tests/configs/test_sparse.template.cfg b/tests/configs/test_sparse.template.cfg
@@ -6,7 +6,7 @@ task=evaluate
 train_location=
 test_location=
 featuresets=[["test_sparse"]]
-learners=['LogisticRegression', 'LinearSVC', 'SVC', 'MultinomialNB', 'DecisionTreeClassifier', 'RandomForestClassifier', 'GradientBoostingClassifier']
+learners=['LogisticRegression', 'LinearSVC', 'SVC', 'MultinomialNB', 'DecisionTreeClassifier', 'RandomForestClassifier', 'GradientBoostingClassifier', 'SGDClassifier']
 suffix=.jsonlines
 
 [Tuning]