为什么 Python 会产生这种奇怪的广播错误？

Question

我有一个表达式 A =​​ B/C，其中 A、B 和 C 都是 1 维 numpy 数组，每个数组有 1258 个元素。 然而 Python 声称它无法将输入数组从 shape (1259) 广播到 shape (1258)。 但是输入数组没有1259的形状。我在除法运算之前打印了所有三个数组的维度，这表明它们的长度都是1258。那么为什么Python会这样呢？

下面是错误消息，（注意顶部所有三个维度的打印）：

print the dimensions of the arrays:
len(close[1:]) 1258
len(close[0:len(close)-1]) 1258
len(close_changes[1:]) 1258
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-6-f1157b8312cc> in <module>
      7 indicator_values_all_stocks = np.zeros(shape=(nr_of_stocks, nr_indicator_values_per_stock))
      8 for i, stock in enumerate(df_open.columns):
----> 9     indicator_values_all_stocks[i] = indicator.change_variance_ratio(df_close[stock], shortLength, longLength)
     10 print(time.time() - start)

~\Desktop\Python Projects Organized\Finance\Indicator Statistics\B.31. Change Variance Ratio\indicator.py in change_variance_ratio(close, shortLength, longLength)
    796     print("len(close_changes[1:])", len(close_changes[1:]))
    797 
--> 798     close_changes[1:] = close[1:]/close[0:len(close)-1]
    799     close_changes[0] = np.NaN
    800 

ValueError: could not broadcast input array from shape (1259) into shape (1258)

编码：

shortLength = 5   
longLength = 50   
nr_of_stocks = len(df_open.columns) 
nr_indicator_values_per_stock = len(df_open)
indicator_values_all_stocks = np.zeros(shape=(nr_of_stocks, nr_indicator_values_per_stock))
for i, stock in enumerate(df_open.columns):
    indicator_values_all_stocks[i] = indicator.change_variance_ratio(df_close[stock], shortLength, longLength)

函数change_variance_ratio：

def change_variance_ratio(close, shortLength, longLength):

    close_changes = np.zeros(len(close))
    
    print("print the dimensions of the arrays:")
    print("len(close[1:])", len(close[1:]))
    print("len(close[0:len(close)-1])", len(close[0:len(close)-1]))
    print("len(close_changes[1:])", len(close_changes[1:]))

    close_changes[1:] = close[1:]/close[0:len(close)-1]   #This line causes the error
    close_changes[0] = np.NaN

    change_variance_ratio = np.zeros(len(close))
    change_variance_ratio[0:longLength] = np.NaN

    for i in range(longLength, len(close)):
        short_val = np.var(np.log(close_changes[i-shortLength:i]))
        long_val = np.var(np.log(close_changes[i-longLength:i]))
        change_variance_ratio[i] = (short_val/long_val)
    
    return change_variance_ratio

Answer 1

问题来自于close熊猫Series的划分。

尝试使用to_numpy()方法修改change_variance_ratio如下：

def change_variance_ratio(close, shortLength, longLength):
    close_array = close.to_numpy()
    close_changes = np.zeros(len(close_array))
    
    print("print the dimensions of the arrays:")
    print("len(close_array[1:])", len(close_array[1:]))
    print("len(close_array[0:len(close_array)-1])", len(close[0:len(close_array)-1]))
    print("len(close_changes[1:])", len(close_changes[1:]))

    close_changes[1:] = close_array[1:]/close_array[0:len(close_array)-1]   #This line causes the error
    close_changes[0] = np.NaN

    change_variance_ratio = np.zeros(len(close_array))
    change_variance_ratio[0:longLength] = np.NaN

    for i in range(longLength, len(close_array)):
        short_val = np.var(np.log(close_changes[i-shortLength:i]))
        long_val = np.var(np.log(close_changes[i-longLength:i]))
        change_variance_ratio[i] = (short_val/long_val)
    
    return change_variance_ratio