嘗試規范化和更新權重時收到運行時警告

Question

我正在嘗試計算粒子過濾器中某些粒子的權重，然后相應地對這些權重進行歸一化。 我的代碼：

def update(particles, weights, landmark, sigma):
    n = 0.0
    for i in range(len(weights)):
        distance = np.power((particles[i][0] - landmark[0]) ** 2 + (particles[i][1] - 
        landmark[1])**2, 0.5)
        likelihood = exp(-(np.power(distance, 2))/2 * sigma ** 2)
        weights[i] = weights[i] * likelihood
        n += weights[i]
        weights += 1.e-30
        if n != 0:
            weights = weights / n

但是，我收到錯誤：/Users/scottdayton/PycharmProjects/Uncertainty Research/particle.py:30: RuntimeWarning: 在 true_divide weights = weights = weights/n /Users/scottdayton/PycharmProjects/Uncertainty Research/particle.py:30 中遇到溢出: RuntimeWarning: true_divide weights = weights / n中遇到的無效值

Answer 1

正如評論中所說，我在您的代碼中添加了括號，但可能還有另一件事。 我覺得您正在嘗試將權重與可能性相乘，然后對結果進行歸一化。 為此，您應該在 2 中切斷循環：

• 校正權重和總和。
• 歸一化總和為一。

我會這樣寫：

def update(particles, weights, landmark, sigma):
    n = 0.0
    # Correction of weights and computation of the sum
    for i in range(len(weights)):
        distance = np.power((particles[i][0] - landmark[0]) ** 2 + (particles[i][1] - 
        landmark[1])**2, 0.5)
        likelihood = np.exp(-(np.power(distance, 2))/(2 * sigma ** 2))
        weights[i] = weights[i] * likelihood + 1.e-30
        n += weights[i]
    # Normalization to sum up to one
    for i in range(len(weights)):
        weights[i] = weights[i] / n