updating a value in a multi index data frame-CodePudding

could someone please explain this behavior to me? this is my dataframe:

df = pd.DataFrame(
    {
    'key1' : ['f1', 'f1', 'f2', 'f2', 'f3'],
    'key2' : ['fm1', 'fm2', 'fm1', 'fm2', 'fm1'],

    'k' : np.random.rand(5),
    'c' : [100, 200, 100, 150, 400],
    })
df.set_index(["key1","key2"],inplace=True)

when I try to change a value:

df.loc["f1","fm1"].loc["k"]=-999

nothing happens. and df.loc["f1","fm1"].loc["k"] still gives the old value I am wondering what I am doing wrong, and how should I achieve this?

CodePudding user response：

Use a tuple and only one loc:

>>> df.loc[("f1", "fm1"), "k"] = -999
>>> df
                    k    c
key1 key2                 
f1   fm1  -999.000000  100
     fm2     0.082072  200
f2   fm1     0.119572  100
     fm2     0.752586  150
f3   fm1     0.947604  400
>>>

The reason your code doesn't work, is because df.loc["f1", "fm1"] references to an object that has no interference with the original df, it returns a completely new dataframe and replaces the k value, but it doesn't replace it in df as well.

Read more about it here on the pandas documentation.

CodePudding user response：

Because you're trying to change the value on a copy of a slice from the dataframe. Read about it here: Why does assignment fail when using chained indexing?

The desired result can be achieved using below code instead:

df["k"].loc["f1", "fm1"] = -999