Transformation of data and its impact on final result


1) Instead of using imputer of sklearn can we not manually impute the columns with their mean/median (fillna function in pandas)? What is the difference here?

2) When a column is transformed (either sq root or logarithmic) is it that during interpretation of result we undo this transformation to understand real meaning? If we are transforming only few columns instead of all, doesnt it distort the final result interpretation?