An author identification system based on recur
Python Makefile Shell
Switch branches/tags
Nothing to show
Clone or download
Failed to load latest commit information.
config Remove some seemingly bad ones from DU config May 24, 2015
mappings Truncate long runs of the same letter (at 5) Apr 13, 2015
.gitignore summarise-config: use a subirectory for tidiness May 13, 2015
LICENSE License (GPLv2), readme stub, gitignore Apr 9, 2015
Makefile update Makefile for updated corpus May 13, 2015 README: add links to papers Dec 13, 2016
calc-score calc-score -S sorts ensembles by commit ID May 18, 2015 Add --colour argument to calc-score and summarise-config May 18, 2015
combine-answers flake8 fixes for combine-answers May 13, 2015
corpus-utils corpus-utils: charmap output is sorted Dec 13, 2016 corpus-utils --write-remapped-text shows the transformed text Dec 13, 2016 make-answers keeps logging to the same old place May 13, 2015
make-control-corpus flake8 fixes for make-control-corpus May 13, 2015
pan-ensemble pan-ensemble --randomise now really does randomise May 24, 2015
pan-test pan-test configuration changes May 18, 2015 Shift balance_results, write_answers into shared module May 19, 2015 Add script suite to summarise good configuration combinations May 13, 2015 calc-score -s shows fixed 0.5 cat1 score May 19, 2015
summarise-config Add --colour argument to calc-score and summarise-config May 18, 2015
summarise-config-old summarise-config is hard-wired for the old corpus; leave it May 13, 2015
test-cat1 A simple test for cat1 calculations May 18, 2015
train-net train-net: improve option defaults (suiting adagrad) May 19, 2015

Author identification using multi-headed recurrent neural networks

This software was developed for the PAN 2015 author identification challenge.

It uses a multi-headed recurrent neural network (RNN) from the recur project. The recur module is in the form of a Python extension, and to build it you need to have various packages listed in the recur README, and the python-dev package. make is supposed to work.

There is a paper about it on the pan site and an expanded/modified/reviewed version at

This software is copyright Douglas Bagnall and is licensed under the GPL, version 2 or greater.