Question

我有 3 种不同产品的时间序列，这些产品在一段时间内已在 4 家不同的商店销售。我想把缺失的数据补上，这样我就有了一个完整的数据集。所有缺失的数据都应该用0代替。

这是生成数据集的代码。 randomtimes 函数是从@abernert 复制的 [https://stackoverflow.com/questions/50165501/generate-random-list-of-timestamps-# in-python][1]

import datetime
import random
import pandas as pd
import numpy as np

random.seed(42)
np.random.seed(42) 

def randomtimes(start, end, n):
    stime = datetime.datetime.strptime(start, '%d-%m-%Y')
    etime = datetime.datetime.strptime(end, '%d-%m-%Y')
    td = etime - stime
    print(td)
    dates = [round(random.random(),1) * td + stime for _ in range(n)]
    return dates

  # set vars
nsp = 5 # nr of days
nd  = 3 # nr of days
ns  = 3 # nr of stores
npr = 2 # nr of products

# generate data
total = nd*ns*npr
s  = random.sample('1'*nd*ns +'2'*nd*ns+'3'*nd*ns, total)# number of stores
p  = random.sample("a"*nd*ns+ "b"*nd*ns, total)
so = list(np.random.choice(range(20,100),total))
stime = '01-02-2000'
etime   = '03-02-2000'
date   = np.array(randomtimes(stime, etime, nsp)).astype('datetime64[D]')

product = []
store   = []
sold    = []

for x in range(1,len(date)+1):
    product.append(s.pop()) 
    store.append(p.pop()) 
    sold.append(so.pop())
    
    
data = {'date':date,
        'product':product,
        'sold':sold,
        'store':store
       }
df = pd.DataFrame(data )
df 

            date product  sold store
0 2000-02-02       3    95     b
1 2000-02-01       1    88     a 
2 2000-02-02       1    81     a
3 2000-02-03       1    66     a
4 2000-02-02       3    88     a

这个结果应该是这样的。

 0 2000-02-01       1    88     a
 1 2000-02-01       2     0     a
 2 2000-02-01       3     0     a
 3 2000-02-01       1     0     b
 4 2000-02-01       2     0     b
 5 2000-02-01       3    95     b

 6 2000-02-02       1    81     a
 7 2000-02-02       2     0     a
 8 2000-02-02       3    88     a
 9 2000-02-02       1     0     b
10 2000-02-02       2     0     b
11 2000-02-02       3     0     b

12 2000-02-03       1    66     a
13 2000-02-03       2     0     a
14 2000-02-03       3     0     a
15 2000-02-03       1     0     b
16 2000-02-03       2     0     b
17 2000-02-03       3     0     b

另外，有没有更好的方法来生成这个玩具数据？

感谢您的帮助。

Answer 1

我建议创建一个包含所有已知值（日期、产品、商店）的第二个数据框，并将所有“已售”值设置为零。之后，您可以遍历现有的“已售出”数据并将它们复制到新的数据框中。

您可以在下面找到一个说明此过程的工作示例。

import datetime
import pandas as pd
import numpy as np

np.random.seed(0)
n_days = 5  # number of data frame rows
products = [1,2,3]  # list of product ids
stores = ["a", "b", "c", "d"]  # list of store ids

def simulate_data():
    data = {
        "date": random_days(n_days),
        "product": [np.random.choice(products) for row in range(n_days)],
        "sold": [np.random.choice(range(100)) for row in range(n_days)],
        "store": [np.random.choice(stores) for row in range(n_days)]
        }
    df = pd.DataFrame(data)
    return df

def random_days(n, daydelta=7):
    """
    return: list of random days
    n: number of days
    daydelta: number of days from today on to choose from
    """
    daydeltas = range(daydelta)
    random_daydeltas = [datetime.timedelta(days=int(np.random.choice(daydeltas))) for _ in range(n)]
    today = datetime.date.today()
    random_days = [today + random_daydelta for random_daydelta in random_daydeltas]
    return random_days

def get_dfMask(df):
    duration = range((max(df.date) - min(df.date)).days)
    unique_dates = [min(df.date) + datetime.timedelta(days=days) for days in duration]
    data = {
        "date": get_filledDates(unique_dates),
        "product": products * len(stores) * len(unique_dates),
        "sold": [0] * len(products) * len(stores) * len(unique_dates),
        "store": get_filledStores(unique_dates)
        }
    df_filled = pd.DataFrame(data)
    return df_filled

def get_filledDates(unique_dates):
    dates_filled = []
    for unique_date in unique_dates: 
        dates_filled.extend([unique_date] * len(products) * len(stores))
    return dates_filled

def get_filledStores(unique_dates):
    stores_filled = []
    for store in stores: stores_filled.extend([store] * len(products))
    stores_filled *= len(unique_dates)
    return stores_filled

def copy_soldValues(source_df, destination_df):
    for row in range(source_df.__len__()):
        position = (destination_df.loc[:, "date"] == source_df.loc[row,"date"]) \
                & (destination_df.loc[:, "product"] == source_df.loc[row,"product"]) \
                & (destination_df.loc[:, "store"] == source_df.loc[row, "store"])
        destination_df.loc[position, "sold"] = source_df.loc[row, "sold"]
    return destination_df

def main():
    df = simulate_data()
    filled_df = get_dfMask(df)
    filled_df = copy_soldValues(df, filled_df)
    print(df)
    print(filled_df)

if __name__ == "__main__":
    main()

对我来说，你的第二个问题很难回答，因为我不知道你对模拟的期望。但是，在我的示例中，我使用了稍微修改的实现。

熊猫根据 2 个变量添加缺失的日期

1 个答案: