CODEWITHSUNDEEP
pythonpandasnumpydictionary
shockwave
shockwave

Reputation: 3272

Python: Calculate cumulative amount in Pandas dataframe over a period of time

Objective: Calculate cumulative revenue since 2020-01-01.

I have a python dictionary as shown below

data = [{"game_id":"Racing","user_id":"ABC123","amt":5,"date":"2020-01-01"},
    {"game_id":"Racing","user_id":"ABC123","amt":1,"date":"2020-01-04"},
    {"game_id":"Racing","user_id":"CDE123","amt":1,"date":"2020-01-04"},
    {"game_id":"DH","user_id":"CDE123","amt":100,"date":"2020-01-03"},
    {"game_id":"DH","user_id":"CDE456","amt":10,"date":"2020-01-02"},
    {"game_id":"DH","user_id":"CDE789","amt":5,"date":"2020-01-02"},
    {"game_id":"DH","user_id":"CDE456","amt":1,"date":"2020-01-03"},
    {"game_id":"DH","user_id":"CDE456","amt":1,"date":"2020-01-03"}]

The same dictionary above looks like this as a table

   game_id   user_id  amt  activity date
  'Racing', 'ABC123', 5,   '2020-01-01'
  'Racing', 'ABC123', 1,   '2020-01-04'
  'Racing', 'CDE123', 1,   '2020-01-04'
  'DH',     'CDE123', 100, '2020-01-03'
  'DH',     'CDE456', 10,  '2020-01-02'
  'DH', '    CDE789', 5,   '2020-01-02'
  'DH',     'CDE456', 1,   '2020-01-03'
  'DH',     'CDE456', 1,   '2020-01-03'

Age is calculated as the difference between transaction date and 2020-01-01. Total Payer count is number of payers in each game.

I'm trying to create a dataframe having the cumulative results for a each day from the day of first transaction to the next day of transaction. eg:for game_id Racing we start with an amount of 5 on 2020-01-01 so Age is 0. on 2020-01-02 the amount is still 5 because we don't have a transaction on that day. on 2020-01-03 the amount is 5. but on 2020-01-04 the amount is 7 because we have 2 transactions on this day.

Expected output

Game    Age    Cum_rev    Total_unique_payers_per_game
Racing  0      5          2
Racing  1      5          2
Racing  2      5          2
Racing  3      7          2
DH      0      0          3
DH      1      15         3
DH      2      117        3
DH      3      117        3

How to use window functions in python like how we use in SQL. Is there any better approach to solve this problem?

Upvotes: 0

Views: 629

Answers (1)

rpanai
rpanai

Reputation: 13437

Here the very complicated part is to fill dates. I used an apply but I'm not sure this is the best way

import pandas as pd

data = [{"game_id":"Racing","user_id":"ABC123","amt":5,"date":"2020-01-01"},
        {"game_id":"Racing","user_id":"ABC123","amt":1,"date":"2020-01-04"},
        {"game_id":"Racing","user_id":"CDE123","amt":1,"date":"2020-01-04"},
        {"game_id":"DH","user_id":"CDE123","amt":100,"date":"2020-01-03"},
        {"game_id":"DH","user_id":"CDE456","amt":10,"date":"2020-01-02"},
        {"game_id":"DH","user_id":"CDE789","amt":5,"date":"2020-01-02"},
        {"game_id":"DH","user_id":"CDE456","amt":1,"date":"2020-01-03"},
        {"game_id":"DH","user_id":"CDE456","amt":1,"date":"2020-01-03"}]

df = pd.DataFrame(data)
# we want datetime not object
df["date"] = df["date"].astype("M8[us]")

# we will need to merge this at the end
grp = df.groupby("game_id")['user_id']\
        .nunique()\
        .reset_index(name="Total_unique_payers_per_game")

# sum amt per game_id date
df = df.groupby(["game_id", "date"])["amt"].sum().reset_index()

# dates from 2020-01-01 till the max date in df
dates = pd.DataFrame({"date": pd.date_range("2020-01-01", df["date"].max())})

# add missing dates
def expand_dates(x):
    x = pd.merge(dates, x.drop("game_id", axis=1), how="left")
    x["amt"] = x["amt"].fillna(0)
    return x

df = df.groupby("game_id")\
       .apply(expand_dates)\
       .reset_index().drop("level_1", axis=1)

df["Cum_rev"] = df.groupby("game_id")['amt'].transform("cumsum")

# this is equivalent as long as data is sorted
# df["Cum_rev"] = df.groupby("game_id")['amt'].cumsum()

# merge unique payers per game
df = pd.merge(df, grp, how="left")

# dates difference
df["Age"] = "2020-01-01"
df["Age"] = df["Age"].astype("M8[us]")
df["Age"] = (df["date"]-df["Age"]).dt.days


# then you can eventually filter
df = df[["game_id", "Age", 
         "Cum_rev", "Total_unique_payers_per_game"]]\
       .rename(columns={"game_id":"Game"})

Upvotes: 1

Related Questions