为Pandas DataFrame列的一行分配值

Question

I am trying to reassign a value in a single row of a column in a Pandas DataFrame.

import pandas as pd
import numpy as np

Here's the DataFrame:

test_df = pd.DataFrame({'range_total' : [3000,3000,3000,3000,3000,3000,0,2000,2000,1000,1000,1000,1000,1000,1000],
    'high_boundary' : [6,6,6,6,6,6,7,9,9,15,15,15,15,15,15],
    'dist_num' : [1197, 142, 142, 1197, 159, 159, 0, 1000, 1000, 398, 50, 50, 398, 50, 50],
    'round_num_sum' : [2996, 2996, 2996, 2996, 2996, 2996, 0, 2000, 2000, 996, 996, 996, 996, 996, 996]})

In my code I subset the DataFrame for each value of high_boundary and find the index(es) of the test_df corresponding to the greatest value of dist_num (picking the first if tied). For this example, I set the index to:

sub_idx = 0 

I can access the value using this (and other similar versions) code:

test_df.ix[(test_df.high_boundary == 6), "dist_num"][sub_idx]

that returns:

1197

But assigning a new value fails:

test_df.ix[(test_df.high_boundary == 6), "dist_num"][sub_idx] = 42
test_df.ix[(test_df.high_boundary == 6), "dist_num"][sub_idx]

which still returns:

1197 

But:

test_df.ix[(test_df.high_boundary == 6), "dist_num"] = 42
test_df.ix[(test_df.high_boundary == 6), "dist_num"]

returns:

0    42
1    42 
2    42
3    42
4    42
5    42
Name: dist_num, dtype: int64

I appreciate any help. This is my first post, because I've always found what I've needed on SO until now. I'm using version 0.14.0.

Answer 1

Sometimes you can get copy of (part of) original dataframe test_df .

Especially if you select elements using [...][...]

So you changed one value in copy, not in oryginal test_df

Try for example this:

test_df["dist_num"].ix(test_df.high_boundary == 6)[sub_idx] = 0

and you should get expected result.

Answer 2

Had similar problems in past. Suggest you work though examples at

http://pandas.pydata.org/pandas-docs/stable/indexing.html

specifically the section http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy will help you.

The editorial explanation is that if you are chain slicing using df[][] constructs you frequently are slicing to series 1st and then to values. it is not feasible for pandas to keep track of your original filters to allow you to write back to the slice.

short answer try using a single operator like ".loc" to do selections that you intend to assign to.

Answer 3

When revisiting this code after a few years, I found that the solution posted above now provides an error (using Pandas version 0.20.1 and Python 2.7.13): TypeError: 'Series' objects are mutable, thus they cannot be hashed . In case anyone else has this problem, I've added a solution below.

To update a single element of a subset of a pd.DataFrame , the index values from the subset were found and then the index corresponding to desired row was used to select the element to update

sub_idx = 0

indices = test_df.loc[test_df.high_boundary == 6,"dist_num"].index
print(test_df.loc[indices[sub_idx],"dist_num"])
# 1197
test_df.loc[indices[sub_idx],"dist_num"] = 0 

print(test_df.loc[indices[sub_idx],"dist_num"])
# 0