This Python program cleans your inbox for you, by learning the criteria you use to decide whether an unread email should be deleted or not (in fact sometimes you're interested in reading or keeping some of them). It uses a SGDClassifier from scikit-learn to do that.
Read my blog post to know the story behind and to learn more details about the machine learning part.
To run this program you will need to install all the required dependencies. You can easily do that with the following command:
sudo pip install -r requirements.txt
You'll also need this library from @charlierguo (which is used to interface with the Gmail account).
Please note that this Python script will only run with Python 2 (because of that last library), so if your system default version is Python 3 you'll need to replace pip with pip2 and run the script with the python2 command.
- The code is a mess (it was written in about 7/8 hours)
- Some parts are running very slow (like email deletion: spawning multiple threads could be a way to go)
- It uses JSON files for >60MB databases (still I had no problems so far)
- Not a lot of error checking
- Sometimes emails are not cached correctly and it ends up fetching them again for no reason, which slows things down even more (this will require a bit of debugging)
- Saves your login details (password included) in plain text
- Tested only on Linux
- Only works with Gmail accounts
- Sometimes weird things happen (not very often, but keep in mind that I've only tested the program with my email account and with the kind of emails my inbox was bloated with)
- You may need to adjust some thresholds (like the "uncertainty threshold" or the "training vs. testing dataset split ratio")
- Cleans your inbox
- It does that based on the criteria you taught it to use
- You can clean thousands of unread emails just by teaching it a few hundreds samples
- Puts emails it's uncertain about in a separate folder on your Gmail account
- It's machine learning powered
- It is quite accurate, it really surprised me
- Training is fast
- You can absolutely do a "safe" run before doing a real one, so you can preview how the classifier is taking its decisions
- Remember to train and test the classifier before doing any safe/real run
- Even if this is just an experiment, feedback is appreciated
- Pull requests are welcome
I take no responsibility if you lose important emails. This software is provided as it is, without any warranty. It's not a final product or an idiot-proof program by any means: it's more of an experiment I created for fun. Use at your own risk.
Although it works only on unread emails (which, because of the fact they're actually unread, probably means you don't care a lot about them), it has the ability to trash your personal emails based on its own decisions. You have been warned.