Create column based on row data when column doesn't exist or column is NaN in pandas

I have a dataframe from OSM data. In this I got everything but the colour column in my area. However in other areas the column may exist. Now I want to create the column if it is missing by providing calculated colors and also want to replace any NaN values with a color code when the column exists but a row has no color value yet.

TLDR: How do I create a colum if needed and otherwise map NaN otherwise?

I already tried just doing:

import random
def setColor(_):
    r = lambda: random.randint(0,255)
    return '#%02X%02X%02X' % (r(),r(),r())



lines.loc[lines['colour'].isnull(),'colour'] = lines["colour"].map(setColor)

However this fails if colour doesnt exist initially.

I could run lines["colour"] = np.nan first but while that works for empty colums this doesn't work for the case when the column already partially exists. So I wonder if there is a better way.

It's not fully clear what you want, but maybe this is close.

Given df1 and df2:

import pandas as pd
import numpy as np
import random

df1 = pd.DataFrame({'Col_01': ['x', 'y', 'z']})
df2 = pd.DataFrame({'Col_01': ['x', 'y', 'z'], 'colour': ['#D30000', '#C21807', '']})

print("df1:\n", df1)
print("df2:\n", df2)

Console output:

df1:
   Col_01
0      x
1      y
2      z
df2:
   Col_01   colour
0      x  #D30000
1      y  #C21807
2      z

With a slight change to your function (removing argument) and looping through all dataframes:

def setColor(): # change: remove the "_" here
    r = lambda: random.randint(0, 255)
    return '#%02X%02X%02X' % (r(),r(),r())

for df in [df1, df2]:
    if "colour" not in df:
        df["colour"] = df.apply(lambda x: setColor(), axis=1)
    else:
        df["colour"] = np.where(df["colour"] == '', setColor(), df["colour"])

print("df1:\n", df1)
print("df2:\n", df2)

Console output:

df1:
   Col_01   colour
0      x  #C0ACB3
1      y  #1FA09E
2      z  #4A35FF
df2:
   Col_01   colour
0      x  #D30000
1      y  #C21807
2      z  #D97652

It's probably self-explanatory, but the loop first looks to see if the colour column exists; if not, it adds it and creates a hex code for each row. Otherwise, if the column exists, it uses np.where() to create a hex code for blank rows, otherwise keeping hex code if it's there.