Python find_step_lengthの例

コード例 #1

def quasi_newton(f, fd, x, max_iterations, precision, callback):
    I = identity(x.size)
    H = I
    x_prev = x
    f_prev = f
    fd_prev = fd

    for i in range(1, max_iterations):
        gradient = fd(x)
        direction = -H * matrix(gradient).T
        direction = squeeze(asarray(direction))

        alpha = find_step_length(f, fd, x, 1.0, direction, c2=0.9)
        x_prev = x
        x = x + alpha * direction

        callback(i, direction, alpha, x)

        if linalg.norm(x - x_prev) < precision:
            break

        s = matrix(x - x_prev).T
        y = matrix(fd(x) - fd(x_prev)).T
        rho = float(1 / (y.T * s))
        H = (I - rho * s * y.T) * H * (I - rho * y * s.T) + rho * s * s.T
    return x

コード例 #2

ファイル: quasi_newton.py プロジェクト: tamland/non-linear-optimization

def quasi_newton(f, fd, x, max_iterations, precision, callback):
  I = identity(x.size)
  H = I
  x_prev = x
  f_prev = f
  fd_prev = fd
  
  for i in range(1, max_iterations):
    gradient = fd(x)
    direction = -H * matrix(gradient).T
    direction = squeeze(asarray(direction))
    
    alpha = find_step_length(f, fd, x, 1.0, direction, c2=0.9)
    x_prev = x
    x = x + alpha*direction
    
    callback(i, direction, alpha, x)
    
    if linalg.norm(x - x_prev) < precision:
      break
    
    s = matrix(x - x_prev).T
    y = matrix(fd(x) - fd(x_prev)).T
    rho = float(1 / (y.T*s))
    H = (I - rho*s*y.T)*H*(I - rho*y*s.T) + rho*s*s.T
  return x

コード例 #3

ファイル: steepest_descent.py プロジェクト: tamland/non-linear-optimization

def steepest_descent(f, fd, x, max_iterations, precision, callback):
  for i in range(0, max_iterations):
    direction = - fd(x)
    alpha = find_step_length(f, fd, x, 1.0, direction, c2=0.9)
    x = x + alpha*direction
    callback(i, direction, alpha, x)
    if linalg.norm(direction) < precision:
      break
  return x

コード例 #4

def steepest_descent(f, fd, x, max_iterations, precision, callback):
  for i in range(0, max_iterations):
    direction = - fd(x)
    alpha = find_step_length(f, fd, x, 1.0, direction, c2=0.9)
    x = x + alpha*direction
    callback(i, direction, alpha, x)
    if linalg.norm(direction) < precision:
      break
  return x

コード例 #5

ファイル: newton.py プロジェクト: tamland/non-linear-optimization

def newton(f, fd, fdd, x, max_iterations, precision, callback):
  for i in range(1, max_iterations):
    gradient = fd(x)
    hessian = fdd(x)
    
    direction = -linalg.solve(hessian, gradient)
    alpha = find_step_length(f, fd, x, 1.0, direction, c2=0.9)
    x_prev = x
    x = x + alpha*direction
    
    callback(i, direction, alpha, x)
    
    if linalg.norm(x - x_prev) < precision:
      break
  return x

コード例 #6

ファイル: newton.py プロジェクト: manasijy/Optimization-Techniques

def newton(f, fd, fdd, x, max_iterations, precision, callback):
    for i in range(1, max_iterations):
        gradient = fd(x)
        hessian = fdd(x)

        direction = -linalg.solve(hessian, gradient)
        alpha = find_step_length(f, fd, x, 1.0, direction, c2=0.9)
        x_prev = x
        x = x + alpha * direction

        callback(i, direction, alpha, x)

        if linalg.norm(x - x_prev) < precision:
            break
    return x

コード例 #7

ファイル: conjugate_gradient.py プロジェクト: manasijy/Optimization-Techniques

def conjugate_gradient(f, fd, x, max_iterations, precision, callback):
    direction = -fd(x)
    gradient = None
    gradient_next = matrix(fd(x)).T
    x_prev = None

    for i in range(1, max_iterations):
        alpha = find_step_length(f, fd, x, 1.0, direction, c2=0.1)
        x_prev = x
        x = x + alpha * direction

        callback(i, direction, alpha, x)

        gradient = gradient_next
        gradient_next = matrix(fd(x)).T

        if linalg.norm(x - x_prev) < precision:
            break

        BFR = (gradient_next.T * gradient_next) / (gradient.T * gradient)
        BFR = squeeze(asarray(BFR))

        direction = -squeeze(asarray(gradient_next)) + BFR * direction
    return x

コード例 #8

ファイル: conjugate_gradient.py プロジェクト: tamland/non-linear-optimization

def conjugate_gradient(f, fd, x, max_iterations, precision, callback):
  direction = -fd(x)
  gradient = None
  gradient_next = matrix(fd(x)).T
  x_prev = None
  
  for i in range(1, max_iterations):
    alpha = find_step_length(f, fd, x, 1.0, direction, c2=0.1)
    x_prev = x
    x = x + alpha*direction
    
    callback(i, direction, alpha, x)
    
    gradient = gradient_next
    gradient_next = matrix(fd(x)).T
    
    if linalg.norm(x - x_prev) < precision:
      break
    
    BFR = (gradient_next.T * gradient_next) / (gradient.T * gradient)
    BFR = squeeze(asarray(BFR))
    
    direction = -squeeze(asarray(gradient_next)) + BFR*direction
  return x