The task is to choose 1000 clients randomly. In the dataset, the number of rows and unique IDs are the same.
I randomly select and then try to select the dataset with choosen IDs:
from numpy.random import choice
ids = choice(df.id, 1000)
df[df.id.isin(ids)]
The result is quite different: size of df[df.id.isin(ids)] is equal to 917. Why aren't there 1000 rows?
The default for np.random.choice() is to sample with replacement, replace=True.
Given a sample size of 1000, it's fairly likely that there will be a few repeats in the resulting sample.
E.g.
>>> import pandas as pd
>>> import numpy as np
>>> np.random.seed(444)
>>> s = pd.Series(list(range(1000)))
>>> ids = np.random.choice(s, 1000)
>>> s[s.isin(ids)].count()
623
>>> ids_no_rep = np.random.choice(s, 1000, replace=False)
>>> s[s.isin(ids_no_rep)].count()
1000
If what you're looking to do is to shuffle the values of that column in-place, you can call np.random.shuffle(df["col"].values) for that column.