Remove rows from a pandas dataframe at random without shuffling dataset

Question:

I’ve got a dataset which needs to omit a few rows whilst preserving the order of the rows. My idea was to use a mask with a random number between 0 and the length of my dataset but I’m not sure how to setup a mask without shuffling the rows around i.e. a method similar to sampling a dataset.

Example: Dataset has 5 rows and 2 columns and I would like to remove a row at random.

Col1  Col2
   A     1
   B     2 
   C     5     
   D     4
   E     0

transforms to:

Col1  Col2
   A     1
   B     2   
   D     4
   E     0

with the third row (Col1='C') omitted by a random choice.

How should I go about this?

Asked By: Black

||
Source

Answers:

The following should work for you. Here I sample remove_n random row_ids from df‘s index. After that df.drop removes those rows from the data frame and returns the new subset of the old data frame.

import pandas as pd
import numpy as np
np.random.seed(10)

remove_n = 1
df = pd.DataFrame({"a":[1,2,3,4], "b":[5,6,7,8]})
drop_indices = np.random.choice(df.index, remove_n, replace=False)
df_subset = df.drop(drop_indices)

DataFrame df:

    a   b
0   1   5
1   2   6
2   3   7
3   4   8

DataFrame df_subset:

    a   b
0   1   5
1   2   6
3   4   8
Answered By: cel

We could sample the frame and sort the index afterwards.

n_remove = 2
df1 = df.sample(n=len(df)-n_remove).sort_index()

Another way is to sort the randomly chosen indices and filter.

keep_idx = np.random.default_rng().choice(len(df), replace=False, size=len(df)-n_remove)
keep_idx.sort()

df1 = df.take(keep_idx)

res

Answered By: cottontail