Applying Groupby with an np.where function to detect a pattern

I've made a function that finds a pattern in a series in a dataframe using np.where. The function finds a series of three <0 values where each is lower than the previous. If the 4th value is higher than the 3rd, the function returns 1.
The function works but I need to use groupby to apply it to all the names in the table.

Here is the working code:

import pandas as pd
import numpy as np
def PFunc1():
    val = np.where((
        (df1['Score'].shift(+3)<0) &
        (df1['Score'].shift(+1)<0) &
        (df1['Score'].shift(+2) < df1['Score'].shift(+3)) & 
        (df1['Score'].shift(+1) < df1['Score'].shift(+2)) & 
        (df1['Score'] > df1['Score'].shift(+1))),1,0)
    return val

df1 = pd.DataFrame()
df1['Name'] = ['A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A',
               'B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B']
df1['Score'] = np.random.randint(-4,3,df1.shape[0])
df1['Pattern'] = PFunc1()
df1.head(50)

When I run the below I get the error: TypeError: unhashable type: 'numpy.ndarray'
Applying the same with a Lambda function results in Nans.

df1['Pattern2'] = df1.groupby('Name')['Score'].apply(PFunc1())

Is this possible with np.where or is a different approach needed?
Many thanks

>The function works but I need to use groupby to apply it to all the names in the table.

It looks like you aren't performing an aggregation; you're performing an item-by-item transformation. Therefore, don't use GroupBy.apply(), use GroupBy.transform().

The transforming function needs to accept a Series as input, so you should modify PFunc1 to accept an argument.

Also, PFunc1 can be slightly streamlined:

• It doesn't require np.where
• You don't need to check if scores.shift(+1) < 0
  • (If the first item is less than 0 and the next two items are even less than the first, then there's no need to check if they are also less than 0.)

import pandas as pd
import numpy as np

def PFunc1(scores):
    return (
        (scores.shift(+3)<0) &
        (scores.shift(+2) < scores.shift(+3)) &
        (scores.shift(+1) < scores.shift(+2)) &
        (scores > scores.shift(+1))
    ).astype(int)


df1 = pd.DataFrame()
df1['Name'] = ['A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A','A',
               'B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B','B']
df1['Score'] = np.random.randint(-4,0,df1.shape[0])

# Process all Scores without respect to Name
df1['Pattern'] = PFunc1(df1['Score'])

# Process Scores for each Name independently
df1['Pattern2'] = df1.groupby('Name')['Score'].transform(PFunc1)