Melt pandas dataframe based on condition-CodePudding

I have a dataframe with the following format

timestamp   ID    Col1    Col2    Col3    Col4    UsefulCol
16/11/2021  1     0.2             0.1             Col3
17/11/2021  1     0.3             0.8             Col3
17/11/2021  2              10                     Col2
17/11/2021  3                     0.1     2       Col4

And I want to "melt" it into this format:

timestamp   ID   Col   Value
16/11/2021  1    Col3  0.1
17/11/2021  1    Col3  0.8
17/11/2021  2    Col2  10
17/11/2021  3    Col4  2

How would I go about this?

Input as dataframe:

from numpy import nan
df = pd.DataFrame({'timestamp': ['16/11/2021', '17/11/2021', '17/11/2021', '17/11/2021'],
                   'ID': [1, 1, 2, 3],
                   'Col1': [0.2, 0.3, nan, nan],
                   'Col2': [nan, nan, 10.0, nan],
                   'Col3': [0.1, 0.8, nan, 0.1],
                   'Col4': [nan, nan, nan, 2.0],
                   'UsefulCol': ['Col3', 'Col3', 'Col2', 'Col4']})

CodePudding user response：

Try making a column with the useful values first:

df['Value'] = df.apply(lambda x: x[x.UsefulCol], axis=1)

timestamp   ID    Col1    Col2    Col3    Col4    UsefulCol    Value
16/11/2021  1     0.2             0.1             Col3         0.1
17/11/2021  1     0.3             0.8             Col3         0.8
17/11/2021  2              10                     Col2         10
17/11/2021  3                     0.1     2       Col4         2

Then, you can drop the columns you wanted to melt:

df.drop(['Col1', 'Col2', 'Col3', 'Col4], axis=1, inplace=True)

timestamp   ID    UsefulCol    Value
16/11/2021  1     Col3         0.1
17/11/2021  1     Col3         0.8
17/11/2021  2     Col2         10
17/11/2021  3     Col4         2

Rename your columns if you need:

df.rename({'UsefulCol':'Col'}, axis=1, inplace=True)

df.columns = [timestamp', 'ID', 'Col', 'Value]

CodePudding user response：

Here is a vectorial solution using a bit of numpy:

import numpy as np

# select columns to pseudo-melt (this could be a manual list cols=['A', 'B', 'C'])
cols = df.filter(regex='^Col').columns

# slice the needed values (they will be on the diagonal) and keep only diagonal
df['Value'] = np.diag(df.filter(regex='^Col').loc[:, df['UsefulCol']].values)

# drop old columns
new_df = df.drop(columns=cols)

output:

    timestamp  ID UsefulCol     Value
0  16/11/2021   1      Col3    0.1000
1  17/11/2021   1      Col3    0.8000
2  17/11/2021   2      Col2   10.0000
3  17/11/2021   3      Col4    2.0000