Update a dataframe in pandas while iterating row by row

Question

You can assign values in the loop using df.set_value:

for i, row in df.iterrows():
    ifor_val = something
    if <condition>:
        ifor_val = something_else
    df.set_value(i,'ifor',ifor_val)

If you don't need the row values you could simply iterate over the indices of df, but I kept the original for-loop in case you need the row value for something not shown here.

update

df.set_value() has been deprecated since version 0.21.0 you can use df.at() instead:

for i, row in df.iterrows():
    ifor_val = something
    if <condition>:
        ifor_val = something_else
    df.at[i,'ifor'] = ifor_val

Pandas DataFrame object should be thought of as a Series of Series. In other words, you should think of it in terms of columns. The reason why this is important is because when you use pd.DataFrame.iterrows you are iterating through rows as Series. But these are not the Series that the data frame is storing and so they are new Series that are created for you while you iterate. That implies that when you attempt to assign tho them, those edits won't end up reflected in the original data frame.

Ok, now that that is out of the way: What do we do?

Suggestions prior to this post include:

1. pd.DataFrame.set_value is deprecated as of Pandas version 0.21
2. pd.DataFrame.ix is deprecated
3. pd.DataFrame.loc is fine but can work on array indexers and you can do better

My recommendation
Use pd.DataFrame.at

for i in df.index:
    if <something>:
        df.at[i, 'ifor'] = x
    else:
        df.at[i, 'ifor'] = y

You can even change this to:

for i in df.index:
    df.at[i, 'ifor'] = x if <something> else y

---

Response to comment

and what if I need to use the value of the previous row for the if condition?

for i in range(1, len(df) + 1):
    j = df.columns.get_loc('ifor')
    if <something>:
        df.iat[i - 1, j] = x
    else:
        df.iat[i - 1, j] = y

A method you can use is itertuples(), it iterates over DataFrame rows as namedtuples, with index value as first element of the tuple. And it is much much faster compared with iterrows(). For itertuples(), each row contains its Index in the DataFrame, and you can use loc to set the value.

for row in df.itertuples():
    if <something>:
        df.at[row.Index, 'ifor'] = x
    else:
        df.at[row.Index, 'ifor'] = x

    df.loc[row.Index, 'ifor'] = x

Under most cases, itertuples() is faster than iat or at.

Thanks @SantiStSupery, using .at is much faster than loc.

You should assign value by df.ix[i, 'exp']=X or df.loc[i, 'exp']=X instead of df.ix[i]['ifor'] = x.

Otherwise you are working on a view, and should get a warming:

-c:1: SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame. Try using .loc[row_index,col_indexer] = value instead

But certainly, loop probably should better be replaced by some vectorized algorithm to make the full use of DataFrame as @Phillip Cloud suggested.

It's better to use lambda functions using df.apply() -

df["ifor"] = df.apply(lambda x: {value} if {condition} else x["ifor"], axis=1)

Well, if you are going to iterate anyhow, why don't use the simplest method of all, df['Column'].values[i]

df['Column'] = ''

for i in range(len(df)):
    df['Column'].values[i] = something/update/new_value

Or if you want to compare the new values with old or anything like that, why not store it in a list and then append in the end.

mylist, df['Column'] = [], ''

for <condition>:
    mylist.append(something/update/new_value)

df['Column'] = mylist

for i, row in df.iterrows():
    if <something>:
        df.at[i, 'ifor'] = x
    else:
        df.at[i, 'ifor'] = y