How would you plugin custom code for classifying account transactions? After a year of importing OFX files from my checking account, I'm still finding the default classifier very inaccurate and learning very little, if at all.

How would I manually specify a rule like:

Code: Select all

if 'wawa' in memo:
    if amount < 20:
        category='food'
    else:
        category='car:fuel'


Ideally, I'd like to be able to define custom features and then plugin a Maximum Entropy or SVM classifier and train them to deduce these rules based on past manual classifications. Is this possible?

So I'm guessing this isn't possible.

Could someone at least give me some pointers on how I might implement this myself (the KMyMoney interaction, not the classifier)?

You should take a look at the QIF import, and the statement import.

Cerin wrote:So I'm guessing this isn't possible.

Could someone at least give me some pointers on how I might implement this myself (the KMyMoney interaction, not the classifier)?

Sorry, just saw that post now.

In case you want to add this, you should do it independent from the data source (QIF, OFX, HBCI, ...). So this would best be done in converter/mymoneystatementreader.[cpp|h], AFAICT from the top of my head.

If you need more help/advise/support, please let us know. Any improvement against the current state of affairs would be cool to have.