Skip to content

Latest commit

 

History

History
87 lines (63 loc) · 3 KB

README.md

File metadata and controls

87 lines (63 loc) · 3 KB
Raw

evernote-classify

Evernote-classify is a fork of Geeknote which adds automatic classification of your Evernote notes into notebooks. At the moment the program only provides recommendations of last five recently updated notes.

Installation

Download the repository:

$ git clone git://github.com/alexksikes/geeknote-classify.git
$ cd geeknote-classify

Install the excellent scikit-learn machine learning library:

$ pip install -U scikit-learn

Example of Usage

First you need to login using geeknote:

$ python geeknote.py login

Then train a classifier:

$ python geeknote.py train

By default "train" will only fetch the last 300 recently updated notes from your account. Use "--counts" to override this behavior. You can also save the fetched notes with the "--save-notes" command (useful for offline testing, see below).

You can now see the recommendation of the classifier:

$ python geeknote.py classify

Checking Everything Works

Two datasets are provided. The first one comes from Teddy Note and the second one from a sample of the 20 newsgroup dataset.

You can directly test these datasets:

$ python classifier.py ./classifier_dataset/teddy.train ./classifier_dataset/teddy.test

You should get something like that:

...
Making a dataset from the notes saved in classifier_dataset/teddy.train.
Training ...
Estimating model performance with 5 folds cross validation ...
Accuracy: 0.93 (+/- 0.02)
Making a dataset for the notes saved in classifier_dataset/teddy.test.
Evaluating the classifier on this dataset.
Showing the suggestions (current notebook >> suggested notebook).
 0 : 07.12.2012  travel >> travel    An Cabo Verde Windsurf Center
 1 : 07.12.2012  debate >> debate    Should U.S. Continue Aid to Pakistan? - Council on Foreign Relations
 2 : 07.12.2012  debate >> debate    Writing From a Global Perspective
 3 : 07.12.2012  useful >> travel    Getting Started
 4 : 23.03.2012  debate >> debate    The Long War Journal - Charts on US Strikes in Pakistan

Additionally you can login into "sandbox_classify" and "sandbox_classify2" (password is the same as login). "sandbox_classify" is the account holding the Teddy Note dataset. "sandbox_classify2" holds the 20 newsgroup dataset.

Future / Todo

  • sync notes instead of downloading
  • 2-way sync of files (gnsync is only from local to remote)
  • not just the 5 updated notes ... more options
  • more datasets to play with
  • better documentation (doctest)
  • use logging instead of print
  • use scikit standard namespace
  • train on the classified notes, suggest on the unclassified ones
  • auto assign to notebooks
  • integrate with existing unit testing framework
  • grid search of best param of classifier
  • feature selection
  • notes with diff kinds of content (not just text)