Question

在此函数中，exs被假定为浮点数列表。它代表了我所有训练示例的列表，每个示例都是一个浮动列表（长度num_vars），代表感知器输入。假设target是浮点列表（长度num_vars），代表目标函数的系数。

def gradDesc(exs, target, num_vars, n=0.5, its=256):
    import random
    weights = []
    # Create and initialize delWeights to 0. Make its size num_vars.
    delWeights = [0.0]*num_vars


    # Initializes the weights to a real number in [-1,1]. Also makes weights
    # contain num_vars entries.
    for i in range(num_vars):
        weights.append(random.uniform(-1,1))

    # To make the printouts look nicer
    print("Iteration\tError")
    print("---------\t-----")

    for i in range(its):

        # Reset delWeights to 0
        for j in range(num_vars):
            delWeights[j] = 0

        for e in exs:

            # Plug e into the current hypothesis and get the output.
            output = test_hypo(weights, e, num_vars)

            print("delWeights: ", delWeights)
            for dw in delWeights:
                print("type(dw): ", type(dw))
                delWeights[dw] = delWeights[dw] + n*(test_hypo(target, e, num_vars) - output)*e[dw]

        for w in weights:
            weights[w] = weights[w] + delWeights[dw]

        # Print out the error every tenth iteration
        if i % 10 == 0:
            print(i + "\t" + train_err(exs, target, weights, num_vars))

    # Print out the final hypothesis
    print(i + "\t" + train_err(exs, target, weights, num_vars))

    return weights

问题是，当我尝试在给定（有限）测试输入的情况下运行此程序时

trainers = 
[[1, 2.7902232015508766, -4.624194135789617], 
[1, -7.964359679418456, 2.1940274082288624], 
[1, 8.445941538761794, -8.86567924774781], 
... other sub-lists following this same format ...]

和

target = [-2, 1, 2]

我得到这个奇怪的输出：

gradDesc(trainers, target, num_vars)
Iteration       Error
---------       -----
delWeights:  [0, 0, 0]
type(dw):  <class 'int'>
type(dw):  <class 'int'>
type(dw):  <class 'int'>
delWeights:  [0.0, 0, 0]
type(dw):  <class 'float'>
Traceback (most recent call last):

  File "<ipython-input-19-97298b385113>", line 1, in <module>
    gradDesc(trainers, target, num_vars)

  File "C:/Users/Me/.spyder-py3/Machine Learning/gradDesc.py", line 107, in gradDesc
    delWeights[dw] = delWeights[dw] + n*(test_hypo(target, e, num_vars) - output)*e[dw]

TypeError: list indices must be integers or slices, not float

所以我的问题是：为什么dw的类型在通过for e in exs循环的第二次迭代中从int变为float？

Answer 1

您是要使用delWeight而不是for i in range(len(delWeights))遍历for dw in delWeights的索引吗？

for dw in delWeights遍历delWeights中的所有值，因此循环的第一次迭代可能为delWeights中的delWeights[dw] + n*(test_hypo(target, e, num_vars) - output)*e[dw]个索引之一分配了浮点数。

Answer 2

delWeights[dw] = delWeights[dw] + n*(test_hypo(target, e, num_vars) - output)*e[dw]

将delWeights[dw]设置为浮点数，因为e[dw]是浮点数。因此，下次您执行for dw in delWeights:循环时，dw是一个浮点数。

使用delWeights的元素作为索引没有意义。如果要遍历列表并获取索引，则应使用enumerate()

for i, dw in enumerate(delWeights):
    delWeights[i] = dw + n*(test_hypo(target, e, num_vars) - output)*e[i]

Python嵌套循环-内部循环迭代器类型更改

2 个答案: