CODEWITHSUNDEEP
pythonpandasdataframepandas-groupby
Sid
Sid

Reputation: 123

Create conditional column using GroupBy

I want to create a new column in my data frame based on grouping variable in a column in the data frame and then check for condition in another column in the data frame.

I have tried to use np.where with pandas pd.groupby to create a Status column in the data frame where I am checking if the next value in the column is greater than the current one based on grouping each of the Sensor_ID's and based on this I am trying to assign if the Status is set to reset or not_reset, however I haven't been successful with the code.

import pandas as pd
df = pd.DataFrame(data = {'Sensor_ID':['A1', 'A1', 'A1', 'A2','A2', 'A2', 'A2', 'A3', 'A3', 'A3', 'A3', 'A3'], 'Reading':[81, 83.5, 87, 90, 81, 82, 85, 78, 79, 78, 80, 78]})
df

   Sensor_ID  Reading
0         A1     81.0
1         A1     83.5
2         A1     87.0
3         A2     90.0
4         A2     81.0
5         A2     82.0
6         A2     85.0
7         A3     78.0
8         A3     79.0
9         A3     78.0
10        A3     80.0
11        A3     78.0

I want to create the following condition below using np.where, but I want to use the Sensor_ID as a grouping variable.

df['Status'] = np.where(df.Reading.shift(-1) > df.Reading, 'not_reset', 'reset')

I used np.where with groupby and transform

df['Status'] = np.where(df.groupby('Sensor_ID')['Reading'].transform(df['Reading'].shift(-1) > df['Reading'], 'not_reset', 'reset'))

TypeError: 'Series' objects are mutable, thus they cannot be hashed

I have also tried to use apply and transform with groupby but I get error:

df['Status'] = df.groupby('Sensor_ID').apply(lambda row: 'not_reset' if row['Reading'].shift(-1) > row['Reading'] else 'reset')

ValueError: The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all(). --> As its comparing the whole series.

and

df['Status'] = df.groupby('Sensor_ID').transform(df['Reading'].shift(-1) > df['Reading'], 'not_reset', 'reset')
TypeError: 'Series' objects are mutable, thus they cannot be hashed

Expected output:

       Sensor_ID  Reading     Status
0             A1     81.0  not_reset
1             A1     83.5  not_reset
2             A1     87.0  not_reset
3             A2     90.0  not_reset
4             A2     81.0      reset
5             A2     82.0  not_reset
6             A2     85.0  not_reset
7             A3     78.0  not_reset
8             A3     79.0  not_reset
9             A3     78.0      reset
10            A3     80.0  not_reset
11            A3     78.0      reset

Upvotes: 3

Views: 2098

Answers (1)

cs95
cs95

Reputation: 403218

You'll need to apply the condition after the grouping IOW, use the result of groupby with np.where).

I would use groupby and diff, it's the same as comparing the shifted-by-1 value. It's as simple as,

np.where(
    df.groupby('Sensor_ID')['Reading'].diff().fillna(1) > 0, 'not reset', 'reset')

array(['not reset', 'not reset', 'not reset', 'not reset', 'reset',
       'not reset', 'not reset', 'not reset', 'not reset', 'reset',
       'not reset', 'reset'], dtype='<U9')

also see here for the initial version of my answer addressing a (now removed) question.

df['Status'] = np.where(
    df.groupby('Sensor_ID')['Reading'].diff().fillna(1) > 0, 'not reset', 'reset')
df

   Sensor_ID  Reading     Status
0         A1     81.0  not reset
1         A1     83.5  not reset
2         A1     87.0  not reset
3         A2     90.0  not reset
4         A2     81.0      reset
5         A2     82.0  not reset
6         A2     85.0  not reset
7         A3     78.0  not reset
8         A3     79.0  not reset
9         A3     78.0      reset
10        A3     80.0  not reset
11        A3     78.0      reset

Upvotes: 2

Related Questions