diff --git a/scikits/learn/glm/benchmarks/bench_lars.py b/scikits/learn/glm/benchmarks/bench_lars.py
deleted file mode 100644
index bb5e83d5c5ee1484a966813250d5cb338b753c22..0000000000000000000000000000000000000000
--- a/scikits/learn/glm/benchmarks/bench_lars.py
+++ /dev/null
@@ -1,22 +0,0 @@
-"""
-Benchmark for the LARS algorithm.
-
-Work in progress
-"""
-
-from datetime import datetime
-import numpy as np
-from scikits.learn import glm
-
-n, m = 100, 50000
-
-X = np.random.randn(n, m)
-y = np.random.randn(n)
-
-if __name__ == '__main__':
-    print "Computing regularization path using the LARS ..."
-    start = datetime.now()
-    alphas, active, path = glm.lars_path(X, y, method='lasso')
-    print "This took ", datetime.now() - start
-
-
diff --git a/scikits/learn/svm/base.py b/scikits/learn/svm/base.py
index 707b0ff62f64dd14b4c02cd8f5b936cdaa97ec21..0cc36bafb42832efc78966dec5152f10f514e408 100644
--- a/scikits/learn/svm/base.py
+++ b/scikits/learn/svm/base.py
@@ -12,7 +12,7 @@ class BaseLib(BaseEstimator):
             self.weight_label = np.asarray(uy, dtype=np.int32, order='C')
             self.weight = np.array([1.0 / np.sum(y==i) for i in uy],
                                    dtype=np.float64, order='C')
-            self.weight *= y.shape[0] / np.sum(self.weight)
+            self.weight *= uy.shape[0] / np.sum(self.weight)
         else:
             self.weight = np.asarray(class_weight.values(),
                                      dtype=np.float64, order='C')
@@ -77,7 +77,7 @@ class BaseLibSVM(BaseLib):
         X : array-like, shape = [n_samples, n_features]
             Training vector, where n_samples is the number of samples and
             n_features is the number of features.
-        y : array, shape = [n_samples]
+        y : array-like, shape = [n_samples]
             Target values (integers in classification, real numbers in
             regression)
         class_weight : dict, {class_label : weight} or "auto"
@@ -311,10 +311,10 @@ class BaseLibLinear(BaseLib):
 
         Parameters
         ----------
-        X : array-like, shape = [nsamples, nfeatures]
-            Training vector, where nsamples in the number of samples and
-            nfeatures is the number of features.
-        y : array, shape = [nsamples]
+        X : array-like, shape = [n_samples, n_features]
+            Training vector, where n_samples in the number of samples and
+            n_features is the number of features.
+        y : array-like, shape = [n_samples]
             Target vector relative to X
         class_weight : dict , {class_label : weight}
             Weights associated with classes. If not given, all classes
diff --git a/scikits/learn/svm/tests/test_svm.py b/scikits/learn/svm/tests/test_svm.py
index dea35aad3232b8b1c12984b8db1a4bb949353567..f29d76395b051527195ad9dad06654352a2638a9 100644
--- a/scikits/learn/svm/tests/test_svm.py
+++ b/scikits/learn/svm/tests/test_svm.py
@@ -256,12 +256,16 @@ def test_auto_weight():
     # compute reference metrics on iris dataset that is quite balanced by
     # default
     X, y = iris.data, iris.target
-    clf = svm.SVC().fit(X, y)
-    assert_almost_equal(metrics.f1_score(y, clf.predict(X)), 0.94, 2)
+    clf = svm.SVC(kernel="linear").fit(X, y)
+    assert_almost_equal(metrics.f1_score(y, clf.predict(X)), 0.99, 2)
 
     # make the same prediction using automated class_weight
-    clf = svm.SVC().fit(X, y, class_weight="auto")
-    assert_almost_equal(metrics.f1_score(y, clf.predict(X)), 0.99, 2)
+    clf_auto = svm.SVC(kernel="linear").fit(X, y, class_weight="auto")
+    assert_almost_equal(metrics.f1_score(y, clf_auto.predict(X)), 0.99, 2)
+
+    # Make sure that in the balanced case it does not change anything
+    # to use "auto"
+    assert_array_almost_equal(clf.coef_, clf_auto.coef_, 6)
 
     # build an very very imbalanced dataset out of iris data
     X_0 = X[y == 0,:]
@@ -277,7 +281,7 @@ def test_auto_weight():
     # fit a model with auto class_weight enabled
     clf = svm.SVC().fit(X_imbalanced, y_imbalanced, class_weight="auto")
     y_pred = clf.predict(X)
-    assert_almost_equal(metrics.f1_score(y, y_pred), 0.99, 2)
+    assert_almost_equal(metrics.f1_score(y, y_pred), 0.92, 2)
 
 
 def test_error():