Home > front end >  Flatten list of dictionaries into dataframe columns
Flatten list of dictionaries into dataframe columns

Time:04-19

I have the following data which contain lists of dictionaries

data= [
 {'Time': 18057610.0,
  'Flux': [{'V0': -1.4209e-15},
   {'V1': 2.7353e-16},
   {'V2': 1.1935e-15},
   {'V3': 1.1624},
   {'V4': -6.1692e-15},
   {'V5': 3.2218e-15}]},
 {'Time': 18057620.4,
  'Flux': [{'V0': 2.4377e-16},
   {'V1': -6.2809e-15},
   {'V2': 1.6456e-15},
   {'V3': 1.1651},
   {'V4': 1.7147e-15},
   {'V5': 9.8872e-16}]},
 {'Time': 18057631.1,
  'Flux': [{'V0': 4.1124e-15},
   {'V1': 1.5598e-15},
   {'V2': -2.325e-16},
   {'V3': 1.1638},
   {'V4': -3.9983e-15},
   {'V5': 4.459e-16}]}]

I want to get something like this:

preferred_df:

                     V0            V1  ...            V4            V5
Time                                    ...                            
18057610.0 -1.420900e-15  2.735300e-16  ... -6.169200e-15  3.221800e-15
18057620.4  2.437700e-16 -6.280900e-15  ...  1.714700e-15  9.887200e-16
18057631.1  4.112400e-15  1.559800e-15  ... -3.998300e-15  4.459000e-16

I came up with the following code which serves the purpose:

df = pd.DataFrame(data).explode('Flux').reset_index(drop=True)
df = df.join(pd.DataFrame(df.pop('Flux').values.tolist())).groupby('Time').sum()

However, I don't want to use groupby and sum(). What are the other ways (dictionary comprehension?) to flatten the "Flux" column without getting the NaN values while flattening the dictionaries and get the preferred_df? I tried json_normalize() but got same NaNs and needed to use groupby() and sum().

CodePudding user response:

Try:

df = pd.DataFrame(
    [{k: v for d2 in d["Flux"] for k, v in d2.items()} for d in data],
    index=[d["Time"] for d in data],
)
print(df)

Prints:

                      V0            V1            V2      V3            V4            V5
18057610.0 -1.420900e-15  2.735300e-16  1.193500e-15  1.1624 -6.169200e-15  3.221800e-15
18057620.4  2.437700e-16 -6.280900e-15  1.645600e-15  1.1651  1.714700e-15  9.887200e-16
18057631.1  4.112400e-15  1.559800e-15 -2.325000e-16  1.1638 -3.998300e-15  4.459000e-16

CodePudding user response:

You could use the built-in collections.ChainMap method in a list comprehension:

from collections import ChainMap
out = pd.DataFrame([d | ChainMap(*d.pop('Flux')) for d in data])

Output:

         Time            V0            V1            V2      V3            V4            V5
0  18057610.0 -1.420900e-15  2.735300e-16  1.193500e-15  1.1624 -6.169200e-15  3.221800e-15
1  18057620.4  2.437700e-16 -6.280900e-15  1.645600e-15  1.1651  1.714700e-15  9.887200e-16
2  18057631.1  4.112400e-15  1.559800e-15 -2.325000e-16  1.1638 -3.998300e-15  4.459000e-16

CodePudding user response:

one option is pd.json_normalize followed by a groupby (it won't be as efficient as the previous options where the wrangling is done within Python):

( pd
.json_normalize(data, 'Flux', ['Time'])
.groupby('Time', as_index = False)
.min()
)

         Time            V0            V1            V2      V3            V4            V5
0  18057610.0 -1.420900e-15  2.735300e-16  1.193500e-15  1.1624 -6.169200e-15  3.221800e-15
1  18057620.4  2.437700e-16 -6.280900e-15  1.645600e-15  1.1651  1.714700e-15  9.887200e-16
2  18057631.1  4.112400e-15  1.559800e-15 -2.325000e-16  1.1638 -3.998300e-15  4.459000e-1
  • Related