Is there a faster method to do a Pandas groupby cumulative mean?

I am trying to create a lookup reference table in Python that calculates the cumulative mean of a Player’s previous (by datetime) games scores, grouped by venue. However, for my specific need, a player should have previously played a minimum of 2 times at the relevant Venue for a 'Venue Preference' cumulative mean calculation.

df format looks like the following:

DateTime	Player	Venue	Score
2021-09-25 17:15:00	Tim	Stadium A	20
2021-09-27 10:00:00	Blake	Stadium B	30

My existing code that works perfectly, but unfortunately is very slow, is as follows:

import numpy as np
import pandas as pd

VenueSum = pd.DataFrame(df.groupby(['DateTime', 'Player', 'Venue'])['Score'].sum().reset_index(name = 'Sum'))
VenueSum['Cumulative Sum'] = VenueSum.sort_values('DateTime').groupby(['Player', 'Venue'])['Sum'].cumsum()
VenueCount = pd.DataFrame(df.groupby(['DateTime', 'Player', 'Venue'])['Score'].count().reset_index(name = 'Count'))
VenueCount['Cumulative Count'] = VenueCount.sort_values('DateTime').groupby(['Player', 'Venue'])['Count'].cumsum()
VenueLookup = VenueSum.merge(VenueCount, how = 'outer', on = ['DateTime', 'Player', 'Venue'])
VenueLookup['Venue Preference'] = np.where(VenueLookup['Cumulative Count'] >= 2, VenueLookup['Cumulative Sum'] / VenueLookup['Cumulative Count'], np.nan)
VenueLookup = VenueLookup.drop(['Sum', 'Cumulative Sum', 'Count', 'Cumulative Count'], axis = 1)

JavaScript
​x
 
import numpy as np
import pandas as pd
​
VenueSum = pd.DataFrame(df.groupby(['DateTime', 'Player', 'Venue'])['Score'].sum().reset_index(name = 'Sum'))
VenueSum['Cumulative Sum'] = VenueSum.sort_values('DateTime').groupby(['Player', 'Venue'])['Sum'].cumsum()
VenueCount = pd.DataFrame(df.groupby(['DateTime', 'Player', 'Venue'])['Score'].count().reset_index(name = 'Count'))
VenueCount['Cumulative Count'] = VenueCount.sort_values('DateTime').groupby(['Player', 'Venue'])['Count'].cumsum()
VenueLookup = VenueSum.merge(VenueCount, how = 'outer', on = ['DateTime', 'Player', 'Venue'])
VenueLookup['Venue Preference'] = np.where(VenueLookup['Cumulative Count'] >= 2, VenueLookup['Cumulative Sum'] / VenueLookup['Cumulative Count'], np.nan)
VenueLookup = VenueLookup.drop(['Sum', 'Cumulative Sum', 'Count', 'Cumulative Count'], axis = 1)
​

I am sure there is a way to calculate the cumulative mean in one step without first calculating the cumulative sum and cumulative count, but unfortunately I couldn’t get that to work.

Answer

IIUC remove 2 groupby by aggregate by sum and size first and then cumulative sum by both columns:

df1 = df.groupby(['DateTime', 'Player', 'Venue'])['Score'].agg(['sum','count'])
df1 = df1.groupby(['Player', 'Venue'])[['sum', 'count']].cumsum().reset_index()
df1['Venue Preference'] = np.where(df1['count'] >= 2, df1['sum'] / df1['count'], np.nan)
df1 = df1.drop(['sum', 'count'], axis=1)
print (df1)
              DateTime Player      Venue  Venue Preference
0  2021-09-25 17:15:00    Tim  Stadium A               NaN
1  2021-09-27 10:00:00  Blake  Stadium B               NaN

JavaScript
 
df1 = df.groupby(['DateTime', 'Player', 'Venue'])['Score'].agg(['sum','count'])
df1 = df1.groupby(['Player', 'Venue'])[['sum', 'count']].cumsum().reset_index()
df1['Venue Preference'] = np.where(df1['count'] >= 2, df1['sum'] / df1['count'], np.nan)
df1 = df1.drop(['sum', 'count'], axis=1)
print (df1)
              DateTime Player      Venue  Venue Preference
0  2021-09-25 17:15:00    Tim  Stadium A               NaN
1  2021-09-27 10:00:00  Blake  Stadium B               NaN
​

Advertisement

Answer