def run_function(f):
prob = cp.Problem(cp.Minimize(f()))
obj_val0 = prob.solve()
status, obj_val1 = ep.solve(prob)
tol = 1e-2
assert_equal(status, OPTIMAL)
assert_less(abs(obj_val1 - obj_val0) / (1 + abs(obj_val0)), tol)
def run_function(f):
prob = cp.Problem(cp.Minimize(f()))
obj_val0 = prob.solve()
status, obj_val1 = ep.solve(prob)
tol = 1e-2
assert_equal(status, OPTIMAL)
assert_less(abs(obj_val1-obj_val0)/(1+abs(obj_val0)), tol)
V = pca(mnist["X"], 50) X = mnist["X"].dot(V) sigma = median_dist(X) y = mnist["Y"].ravel() ytest = mnist["Ytest"].ravel() # Random features n = 4000 W = np.random.randn(X.shape[1], n) / sigma b = np.random.uniform(0, 2 * np.pi, n) X = np.cos(X.dot(W) + b) Xtest = np.cos(mnist["Xtest"].dot(V).dot(W) + b) # Parameters m, n = X.shape k = 10 Theta = cp.Variable(n, k) lam = 10 # Form problem with CVXPY and solve with Epsilon f = ep.multiclass_hinge_loss(Theta, X, y) + lam * cp.sum_squares(Theta) prob = cp.Problem(cp.Minimize(f)) ep.solve(prob, verbose=True) # Get solution and compute train/test error Theta0 = np.array(Theta.value) print "Train error:", error(np.argmax(X.dot(Theta0), axis=1), y) print "Test error:", error(np.argmax(Xtest.dot(Theta0), axis=1), ytest)
