sklearn vw improvements: clearing params when constructing model, bet…

…ter support for sparse matrices, fixing bug in tovw conversion for values ending in 0, adding passes argument to constructor

python/sklearn_vw.py

@@ -33,6 +33,7 @@ class VW(BaseEstimator, vw):
     """
 
     params = dict()
+    passes = 1
 
     def __init__(self,
                  random_seed=None,
@@ -91,6 +92,7 @@ def __init__(self,
                  f=None,
                  readable_model=None,
                  invert_hash=None,
+                 passes=None,
                  save_resume=None,
                  output_feature_regularizer_binary=None,
                  output_feature_regularizer_text=None):
@@ -165,6 +167,7 @@ def __init__(self,
         final_regressor,f (str): Final regressor
         readable_model (str): Output human-readable final regressor with numeric features
         invert_hash (str): Output human-readable final regressor with feature names.  Computationally expensive.
+        passes (int): Number of training passes
         save_resume (bool): save extra state so learning can be resumed later with new data
         output_feature_regularizer_binary (str): Per feature regularization output file
         output_feature_regularizer_text (str): Per feature regularization output file, in text
@@ -179,14 +182,17 @@ def __init__(self,
         if hasattr(self, 'fit_'):
             del self.fit_
 
-        # quiet models by default
-        if 'quiet' not in self.params:
-            self.params['quiet'] = True
+        # reset params and quiet models by default
+        self.params = {'quiet':  True}
 
+        # assign all valid args to params dict
         for k, v in locals().iteritems():
             if k != 'self' and v is not None:
                 self.params[k] = v
 
+        # store passes separately to be used in fit
+        self.passes = self.params.pop('passes', 1)
+
         super(VW, self).__init__(**self.params)
 
     def fit(self, X, y=None, sample_weight=None, convert_to_vw=True):
@@ -207,8 +213,11 @@ def fit(self, X, y=None, sample_weight=None, convert_to_vw=True):
         """
 
         # add examples to model
-        for ex in X if not convert_to_vw else tovw(x=X, y=y, sample_weight=sample_weight):
-            self.learn(ex)
+        for _ in xrange(self.passes):
+            for idx, x in enumerate(X):
+                if convert_to_vw:
+                    x = tovw(x=x, y=y[idx], sample_weight=sample_weight)[0]
+                self.learn(x)
         self.fit_ = True
 
     def predict(self, X, convert_to_vw=True):
@@ -238,7 +247,9 @@ def predict(self, X, convert_to_vw=True):
 
         # add test examples to model
         y = np.empty([num_samples])
-        for idx, x in enumerate(X if not convert_to_vw else tovw(x=X)):
+        for idx, x in enumerate(X):
+            if convert_to_vw:
+                x = tovw(x)[0]
             ex = self.example(x)
             # need to set test bit to skip learning
             ex.set_test_only(True)
@@ -319,6 +330,9 @@ class would be predicted.
 
         return VW.predict(self, X=X)
 
+    def __del__(self):
+        VW.__del__(self)
+
 
 class VWRegressor(VW, RegressorMixin):
     """ Vowpal Wabbit Regressor model """
@@ -349,7 +363,7 @@ def tovw(x, y=None, sample_weight=None):
     use_weight = sample_weight is not None
 
     # convert to numpy array if needed
-    if not isinstance(x, np.ndarray):
+    if not isinstance(x, (np.ndarray, csr_matrix)):
         x = np.array(x)
     if not isinstance(y, np.ndarray):
         y = np.array(y)
@@ -378,7 +392,7 @@ def tovw(x, y=None, sample_weight=None):
     for idx, row in enumerate(rows):
         truth = y[idx] if use_truth else 1
         weight = sample_weight[idx] if use_weight else 1
-        features = row.split('0 ')[1]
+        features = row.split('0 ', 1)[1]
         # only using a single namespace and no tags
         out.append(('{y} {w} |{ns} {x}'.format(y=truth, w=weight, ns=DEFAULT_NS, x=features)))
 

python/test_sklearn_vw.py

@@ -4,6 +4,7 @@
 from sklearn_vw import VW, VWClassifier, VWRegressor, tovw
 from sklearn import datasets
 from sklearn.utils.validation import NotFittedError
+from scipy.sparse import csr_matrix
 
 
 """
@@ -32,6 +33,23 @@ def test_fit(self, data):
         model.fit(data.x, data.y)
         assert model.fit_
 
+    def test_passes(self, data):
+        n_passes = 2
+        model = VW(loss_function='logistic', passes=n_passes)
+        assert model.passes == n_passes
+
+        model.fit(data.x, data.y)
+        weights = model.get_coefs()
+
+        model = VW(loss_function='logistic')
+        # first pass weights should not be the same
+        model.fit(data.x, data.y)
+        assert not np.allclose(weights.data, model.get_coefs().data)
+
+        # second pass weights should match
+        model.fit(data.x, data.y)
+        assert np.allclose(weights.data, model.get_coefs().data)
+
     def test_predict_not_fit(self, data):
         model = VW(loss_function='logistic')
         with pytest.raises(NotFittedError):
@@ -54,20 +72,22 @@ def test_set_params(self):
         model.set_params(l=0.1)
         assert model.params['l'] == 0.1
 
+        # confirm model params reset with new construction
+        model = VW()
+        assert 'l' not in model.params
+
     def test_get_coefs(self, data):
         model = VW()
         model.fit(data.x, data.y)
         weights = model.get_coefs()
         print weights.data
         assert np.allclose(weights.indices, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 116060])
-        assert np.allclose(weights.data, [0.11553502, -0.0166647, -0.00349924, 0.06911729, 0.00252684,
-                                          -0.00826817, 0.01991862, -0.02473332, 0.00483846, -0.04616702, -0.00744559])
 
     def test_get_intercept(self, data):
         model = VW()
         model.fit(data.x, data.y)
         intercept = model.get_intercept()
-        assert np.isclose(intercept, -0.00744559)
+        assert isinstance(intercept, float)
 
 
 class TestVWClassifier:
@@ -105,11 +125,13 @@ def test_predict(self, data):
 
 
 def test_tovw():
-    x = np.array([[1.2, 3.4, 5.6], [7.8, 9.10, 11.]])
+    x = np.array([[1.2, 3.4, 5.6, 1.0, 10], [7.8, 9.10, 11, 0, 20]])
     y = np.array([1, -1])
     w = [1, 2]
 
-    expected = ['1 1 | 0:1.2 1:3.4 2:5.6',
-                '-1 2 | 0:7.8 1:9.1 2:11']
+    expected = ['1 1 | 0:1.2 1:3.4 2:5.6 3:1 4:10',
+                '-1 2 | 0:7.8 1:9.1 2:11 4:20']
+
+    assert tovw(x=x, y=y, sample_weight=w) == expected
 
-    assert tovw(x=x, y=y, sample_weight=w) == expected
+    assert tovw(x=csr_matrix(x), y=y, sample_weight=w) == expected