[hw1] Reporting results

1 files changed, 7 insertions, 7 deletions

diff --git a/hw1/main.py b/hw1/main.py
index 3813fa7..ae48050 100644
--- a/hw1/main.py
+++ b/hw1/main.py
@@ -53,11 +53,11 @@ def train_qr(X, y):
 
 def log_posterior(w, X, y, sigma=1.0, tau=10):
     return (1 / (2 * sigma) * np.linalg.norm(y - np.dot(X, w)) ** 2
-            + 1 / (2 * tau) * np.linalg.norm(w) ** 2)
+            + tau / 2 * np.linalg.norm(w) ** 2)
 
 
 def gradient(w, X, y, sigma=1.0, tau=10):
-    return (- 1 / sigma * np.dot(X.T, y - np.dot(X, w)) + 1 / tau * w)
+    return (- 1 / sigma * np.dot(X.T, y - np.dot(X, w)) + tau * w)
 
 
 def verify_gradient(eps=1e-10):
@@ -88,9 +88,7 @@ def train_lbfgs(X, y):
                     options={'maxiter': 100}).x
 
 
-def map_features(X, d):
-    A = np.random.normal(0, 1, size=(d, X.shape[1]))
-    b = np.random.uniform(0, 2 * pi, d)
+def map_features(A, b, X):
     return np.cos(np.dot(A, X.T).T + b)
 
 
@@ -110,8 +108,10 @@ if __name__ == "__main__":
     lqr = []
     llbfgs = []
     for d in [100, 200, 400, 600]:
-        X = map_features(X_train, d)
-        X_t = map_features(X_test, d)
+        A = np.random.normal(0, 1, size=(d, X_train.shape[1]))
+        b = np.random.uniform(0, 2 * pi, d)
+        X = map_features(A, b, X_train)
+        X_t = map_features(A, b, X_test)
 
         t_start = time.time()
         wqr = train_qr(X, y_train)