On Tue, Jun 10, 2014 at 8:55 PM, Peter Ross <[email protected]> wrote:
> On 10 June 2014 19:31, Stefano Zacchiroli wrote: > > On Fri, May 30, 2014 at 09:54:39AM +0200, Jostein Berntsen wrote: > >> I like to use reckon: > > > > Has anyone here used both reckon and icsv2ledger, and fancy posting a > > brief comparison of the two? > > > > I've just integrated icsv2ledger in my accounting work-flow. I've > > managed to make it do what I want, but I've also found it a bit rough > > around the edges --- I can elaborate more if people on this list are > > interested. > > > I'm interested as a very part-time developer of icsv2ledger. > > It works well for what I want to do which is to enter 10-20 > transactions a week and get the right payee and account most of the > time, and as a developer, writing a regex is not difficult for me to > do. > When you say, "get the right account," this assumes a lot of context, this assumes that the problem is that there is a missing account to guess, a modest subproblem that I call "categorization." For example, such as would be the case for importing credit card transaction history. The more general problem is that of importing any data from any file may not require solving this. About half of the importing that I personally do does not involve any kind of categorization, e.g. importing historical data from my trading account, and I find the rest to be so easy and to consume so little time that I don't even think of it as a real problem (if I'd say I spend more than 10 minutes every two weeks categorizing transactions I'd be exaggerating). It's a bite-size toy problem to play with to implement a little learning classifier on. More important is avoiding errors, and to that extent, an explicit configuration is what I favor. Something like you find in this section of the Ledgerhub design doc: https://docs.google.com/document/d/11u1sWv7H7Ykbc7ayS4M9V3yKqcuTY7LJ3n1tgnEN2Hk/edit#heading=h.83939k4es7n For each source of document, you create an importer which has a fixed configuration of accounts that are specific to you. The importer does no guessing and generates transactions to the accounts you instantiated it with. It's not clear that one could or would want to learn these. Your configuration changes very little, and you do need to write a small script to configure your importers. Also more important IMO is, given a file, categorizing where it comes from and what kind of importer code to run on it. I call this "identification" in the LedgerHub doc. What I really like is the auto-completion, and what I find painful is > when the auto-completion doesn't work as expected. > > Reckon sounds like a very interesting project, and I really like the > idea that it "learns" account name to use. > > Someone correct me if I'm wrong, but it doesn't learn payee names, but > just uses the name as is from the csv file. > > That personally would drive me nuts, as payee names here in Australia > contain a lot of extraneous information. > The problem of extracting meaningful payee names from the crap that lives in those downloadable files is an interesting one. It might be interesting to join efforts and create a corpus of such payee names to work from. In the US and Canada, many of the payee names have the city and name in them: "BARNES&NOBLE BKSTRE #2000NEW YOR" "BACO MERCAT LOS AN" "SAQ23220 MONT-ROYAL OU -- MONTREAL QC" To me reckon sounds also a more friendly project for people who have > never heard of regexs. > It is a reasonable assumption to me that anybody using a custom computer language to do double-entry accounting will, at least to some extent, be a bit of a nerd. -- --- You received this message because you are subscribed to the Google Groups "Ledger" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
