import pandas as pd
import numpy as np

df = pd.DataFrame.from_dict({'date':[1,2,3,4,5,6,7,8,9,10] ,'open':[4,5,3,4,5,6,7,8,9,10],'close':[4,5,6,7,8,1,2,9,10,11],'stock':['A']*5   ['B']*5})

df['flag'] = np.select([df['close']>df['open'],df['close']<df['open']],['up','down'],default='flat')
df

I tried the following. None of them works. They all give me "No numeric types to aggregate" error

# flag if previous 3 days (t-2,t-1, and t) are all increase for each stock

df['3days_up'] = df.groupby('stock')['flag'].rolling(3).apply(lambda x: 'Yes' if all(x['flag']=='up') else 'No')
df['3days_up'] = df.groupby('stock')[['flag']].rolling(3).apply(lambda x: 'Yes' if all(x['flag']=='up') else 'No')
df['3days_up'] = df.groupby('stock').rolling(3).apply(lambda x: 'Yes' if all(x['flag']=='up') else 'No')

Expected output:

CodePudding user response：

Convert up value to True and others to False as starting point:

df['3days_up'] = np.where(df.assign(is_up=df['flag'] == 'up')
                            .groupby('stock').rolling(3)['is_up']
                            .sum() >= 3, 'Yes', 'No')
print(df)

# Output
   date  open  close stock  flag 3days_up
0     1     4      4     A  flat       No
1     2     5      5     A  flat       No
2     3     3      6     A    up       No
3     4     4      7     A    up       No
4     5     5      8     A    up      Yes
5     6     6      1     B  down       No
6     7     7      2     B  down       No
7     8     8      9     B    up       No
8     9     9     10     B    up       No
9    10    10     11     B    up      Yes