blaecksprutte

blaecksprutte is a tagging script, that works on notmuch databases. you have to train blaecksprutte on your notmuch database, with already tagged emails. from this corpus blaecksprutte learns for which emails you use which tags and tags your mails for you. for now, blaecksprutte can only operate on the default notmuch database on your system. configuration is not possible.

dependencies

blaecksprutte uses scikit-learn and the notmuch python api. make sure, you have installed both of them. also the progressbar package of python is used.

training

to train your personal email tagger use:

python blaecksprutte.py train

the trained model is stored in your maildir in 'blaecksprutte.db'

validating

you can check, how well the classifier can perform on your mails. when you use:

python blaecksprutte.py validate

the classifier is trained on 0.6 of your mails and tested on the other 0.4. a classification report with precision, recall and f1-score for every tag is printed to stdout.

optimizing the hyperparameters

blaecksprutte searches 90 combinations of 3 hyperparameters for the best fitting parameters on your dataset. this is done automatically before your first training run. you can choose the best parameters manually with:

python blaecksprutte.py optimize

tagging your emails

blaecksprutte searches for all emails containing the 'new' tag and calculates probable tags for these emails. the new tag is removed. for tagging your emails use:

python blaecksprutte.py tag

you should be able to add to your notmuch command line as in for example:

watch -n 180 "mbsync -a && notmuch new && python blaecksprutte.py tag"

contributing and improving

please contact me or write me issuse and send patches. input of any kind is welcome :)