Categorizer (a PragueHacks 2016 project)

Simple classification engine for government/municipality documents built with TensorFlow Documents are tagged based on occurrence of certain words and other characteristics of a document.

This project is a prototype for

• mapasamospravy.cz
• edesky.cz

Built during Prague Hacks 2016

Setup

Requirements:

• bash
• python 2.7
• numpy
• tensorflow 0.10.0 (does not work with 0.11.0rc0 due to tensorflow/tensorflow#4715)

sudo pip install numpy
export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.10.0-cp27-none-linux_x86_64.whl
sudo pip install --upgrade $TF_BINARY_URL

Run

Prepare data:

1. copy tagged content files to ./input
2. copy feature vector to features.csv
3. export CATS=`cat cats.txt
4. bash generate-all.sh features.csv $CATS

Train DNN

1. python train.py $CATS

Run classification on new data

1. python predict.py features.csv $CATS output.csv