Setting cell values based on computation of all the rows but one

During the preprocessing of some data, I need to remove some outliers. Due to the nature of the application, I cannot remove the data points themselves, so I want to replace them with the maximum of the other data points within some range. For instance, assume the following toy example:

import pandas as pd
from scipy import stats

df = pd.DataFrame({
    "Name": ["A", "A", "A", "A", "B", "B", "B", "B"],
    "Value": [1, 2, 30, 4, 10, 200, 30, 40],
    "Class": ["S", "S", "S", "S", "X", "X", "X", "X"]
})

Now, let's modify the points that are far from one standard deviation (usually, we do it at 3x the standard deviation, or 99.8% percentile. Here, it is just one std as an example):

df[["zscore"]] = (
    df.groupby(["Name"])
    [["Value"]]
    .transform(lambda x : stats.zscore(x, ddof=1))
)

That gives us something like:

  Name  Value Class    zscore
0    A      1     S -0.593976
1    A      2     S -0.521979
2    A     30     S  1.493940
3    A      4     S -0.377985
4    B     10     X -0.685248
5    B    200     X  1.484705
6    B     30     X -0.456832
7    B     40     X -0.342624

Now, I want to replace all values with zscore >= 1.0 to obtain the following table:

  Name  Value Class    zscore
0    A      1     S -0.593976
1    A      2     S -0.521979
2    A      4     S  1.493940
3    A      4     S -0.377985
4    B     10     X -0.685248
5    B     40     X  1.484705
6    B     30     X -0.456832
7    B     40     X -0.342624

Note that on index 2, Value is changed from 30 to 4. In index 5, Value changes from 200 to 40.

Now, my data frame is big (78M+ lines), and I want to do it using the most efficient, but still short code. I tried this, but it doesn't work:

indices = df["zscore"] > 1.0

df.loc[indices] = (
    df[~indices]
    .groupby("Name")
    .max("Value")
)

which give me

  Name  Value Class    zscore
0    A    1.0     S -0.593976
1    A    2.0     S -0.521979
2  NaN    NaN   NaN       NaN
3    A    4.0     S -0.377985
4    B   10.0     X -0.685248
5  NaN    NaN   NaN       NaN
6    B   30.0     X -0.456832
7    B   40.0     X -0.342624

So, what is the right way to do it, keeping it short and fast?

Of course, I can do it a little bit more verbose (I don't know whether it is the fastest way):

for name, group in df.groupby("Name"):
    indices = group["zscore"] > 1.0
    df.loc[group[indices].index, ["Value"]] = group[~indices][["Value"]].max()[0]

which produces the results I want:

  Name  Value Class    zscore
0    A      1     S -0.593976
1    A      2     S -0.521979
2    A      4     S  1.493940
3    A      4     S -0.377985
4    B     10     X -0.685248
5    B     40     X  1.484705
6    B     30     X -0.456832
7    B     40     X -0.342624

Thanks for your help.

Assuming you have sufficient memory to handle dataset, you can first mask the values in the Value column where zscore is > 1 then group the masked column by Name and transform with max to broadcast the max value per group

m = df['zscore'] > 1
df.loc[m, 'Value'] = df['Value'].mask(m).groupby(df['Name']).transform('max')

  Name  Value Class    zscore
0    A      1     S -0.593976
1    A      2     S -0.521979
2    A      4     S  1.493940
3    A      4     S -0.377985
4    B     10     X -0.685248
5    B     40     X  1.484705
6    B     30     X -0.456832
7    B     40     X -0.342624