Question

我有一个数据框df1：

Date_1     Date_2       i_count c_book
01/09/2019  02/08/2019  2       204
01/09/2019  03/08/2019  2       211
01/09/2019  04/08/2019  2       218
01/09/2019  05/08/2019  2       226
01/09/2019  06/08/2019  2       234
01/09/2019  07/08/2019  2       242
01/09/2019  08/08/2019  2       251
01/09/2019  09/08/2019  2       259
01/09/2019  10/08/2019  3       269
01/09/2019  11/08/2019  3       278
01/09/2019  12/08/2019  3       288
01/09/2019  13/08/2019  3       298
01/09/2019  14/08/2019  3       308
01/09/2019  15/08/2019  3       319
01/09/2019  16/08/2019  4       330
01/09/2019  17/08/2019  4       342
01/09/2019  18/08/2019  4       354
01/09/2019  19/08/2019  4       366
01/09/2019  20/08/2019  4       379
01/09/2019  21/08/2019  5       392
01/09/2019  22/08/2019  5       406
01/09/2019  23/08/2019  6       420
01/09/2019  24/08/2019  6       435
01/09/2019  25/08/2019  7       450
01/09/2019  26/08/2019  8       466
01/09/2019  27/08/2019  9       483
01/09/2019  28/08/2019  10      500
01/09/2019  29/08/2019  11      517
01/09/2019  30/08/2019  12      535
01/09/2019  31/08/2019  14      554

我想基于i_count扩展数据集。 i_count是要复制的行数。因此可以说i_count = 2是否暗示需要为同一行复制2行。

此外，我想创建一个新列c_book_i，这样c_book应该在数据集中的条目内划分。例如，如果i_count = 2表示新数据帧应具有2个条目，而c_book_i应具有2个条目，使得sum(c_book_i) = c_book。最后一个约束是我想在所有情况下都拥有c_book_i > 10。

到目前为止：

def f(x):
    i = np.random.random(len(x))
    j = i/sum(i) * x
    return j

joined_df2 = df1.reindex(df1.index.repeat(df1['i_count']))
joined_df2['c_book_i'] = joined_df2.groupby(['Date_1','Date_2'])['c_book'].transform(f)

这为我提供了相同的东西，但是没有检查c_book应该大于10。很多值小于10。

任何人都可以提供帮助。

谢谢

Answer 1

基于solution

def f(x):
    total = x.iloc[0].astype(int)
    minimum = 10
    dividers = sorted(random.sample(range(minimum, total-minimum, minimum), len(x) - 1))
    return [a - b for a, b in zip(dividers + [total], [0] + dividers)]

工作原理。设总数为12，我们希望将其最小分为2，分成4个部分。我们使用步骤2 => [2,4,6,8,10]得到2到12-2的范围。然后获得任意3个数字2,4,8并添加边框，因此[0,2,4,8,12]。现在，该列表[2,2,4,4]之间的差异将为12（边界之间的差异），且两者之和不得小于2

Answer 2

那又怎么样：

def distribute_randomly(array):

    # This is the minimum to give each:
    minimum = 10

    # This means we have to reserve this amount:
    min_value_sum = len(array)*minimum

    # The rest we can distribute:
    to_distribute = array.sum() - min_value_sum

    # Get random values that all sum up to 1:
    random_values = numpy.random.rand(len(array))
    random_values = random_values/random_values.sum()

    # Return the minimum + a part of what is left to distribute
    return random_values*to_distribute + minimum

# Expand rows based on length of i_count:
df1 = df1.join(df1['i_count'].apply(lambda x: range(x)).explode().rename('dummy'))

# transform cbook_ to randomize
df1['c_book_2'] = df1.groupby('i_count')['c_book'].transform(distribute_randomly)

# Finally make sure they are not below 10:
df1['c_book_i'] = df1['c_book_2'].where(df1['c_book_2']>10, 10)

# If needed:
df1 = df1.reset_index()

编辑：添加了“随机”分发功能。

根据列值展开数据集

2 个答案: