svm cross validation

code/stat159lambda/classification/random_forest/rf.py

@@ -9,7 +9,7 @@ def __init__(self,
                  y,
                  max_features='auto',
                  depth=None,
-                 n_estimators=400):
+                 n_estimators=300):
         self.model = RandomForestClassifier(max_features=max_features,
                                             n_estimators=n_estimators,
                                             max_depth=None,

code/stat159lambda/classification/random_forest/rf_cross_validate.py

@@ -5,7 +5,7 @@
 from stat159lambda.classification import design_matrix as dm
 from stat159lambda.classification.random_forest import rf
 from stat159lambda.classification import partition_volumes as pv
-from stat159lambda.config import REPO_HOME_PATH
+from stat159lambda.config import REPO_HOME_PATH, NUM_VOXELS
 from stat159lambda.linear_modeling import linear_modeling as lm
 from stat159lambda.utils import data_path as dp
 

code/stat159lambda/classification/svm/svm.py

@@ -25,7 +25,7 @@ class Classifier:
     -------
     None
     """
-    def __init__(self, X, y, C=1.0, kernel='rbf', degree=2):
+    def __init__(self, X, y, C=.001, kernel='linear', degree=2):
         if kernel == 'linear':
             self.model = LinearSVC(C=C)
         else:
@@ -35,30 +35,30 @@ def __init__(self, X, y, C=1.0, kernel='rbf', degree=2):
 
 
     def train(self):
-    """
-    Classifier method that fits the SVM model according to the given training
-    data X
+        """
+        Classifier method that fits the SVM model according to the given training
+        data X
 
-    Parameters
-    ----------
-    None
+        Parameters
+        ----------
+        None
 
-    Returns
-    -------
-    self : object
-    """
+        Returns
+        -------
+        self : object
+        """
         self.model.fit(self.X, self.y)
 
     def predict(self, new_data):
-    """
-    Performs classification on samples in new_data
+        """
+        Performs classification on samples in new_data
 
-    Parameters
-    ----------
-    new_data : array
+        Parameters
+        ----------
+        new_data : array
 
-    Returns
-    -------
-    pred: array that contains class labels for samples in new_data
-    """
+        Returns
+        -------
+        pred: array that contains class labels for samples in new_data
+        """
         return self.model.predict(new_data)

code/stat159lambda/classification/svm/svm_cross_validate.py

@@ -0,0 +1,35 @@
+from __future__ import print_function, division
+import numpy as np
+from sklearn.cross_validation import KFold
+from sklearn.metrics import accuracy_score
+from stat159lambda.classification import design_matrix as dm
+from stat159lambda.classification.svm import svm
+from stat159lambda.classification import partition_volumes as pv
+from stat159lambda.config import REPO_HOME_PATH, NUM_VOXELS
+from stat159lambda.linear_modeling import linear_modeling as lm
+from stat159lambda.utils import data_path as dp
+
+
+voxels_sorted_by_t_statistic = lm.VoxelExtractor(1, 'int-ext').t_stat()
+# num_features_values = range(100, NUM_VOXELS/100, 100)
+# for num_features in num_features_values:
+voxel_feature_indices = voxels_sorted_by_t_statistic[:1000]
+design_matrix = dm.DesignMatrix(dp.get_smoothed_2d_path(1, 4),
+    pv.get_train_indices(), voxel_feature_indices)
+
+X_train = design_matrix.get_design_matrix()
+y_train = np.array(design_matrix.get_labels())
+
+cv_accuracies = []
+for train, test in KFold(len(X_train), 5):
+	X_cv_train = X_train[train, :]
+	y_cv_train = y_train[train]
+	X_cv_test = X_train[test, :]
+	y_cv_test = y_train[test]
+	model = svm.Classifier(X_cv_train, y_cv_train)
+	model.train()
+	y_predicted = model.predict(X_cv_test)
+	cv_accuracies.append(accuracy_score(y_predicted, y_cv_test))
+
+print(np.mean(cv_accuracies))
+