SCPInstructions.txt

The SCP (Submodular Contextual Policy) algorithm is designed for domains that involve predicting a list of options. Typical applications include news recommendation, robotic trajectory selection and document summarization, etc. These applications typically require a nice balance between diversity and quality. SCP achieves this via leveraging online submodular optimization: with a single no-regret online learner, we can compete with an optimal sequence of predictions. 

The algorithm ensures both computational and data efficiency. We have released our code on Github. You may try to play with it. 

https://github.com/robinzhoucmu/SCP_ListPrediction

For algorithmic details, please see the paper.

http://arxiv.org/pdf/1305.2532.pdf





This is a instruction file for how to use SCP (Submodular Contexual Policy) code for list prediction. 
For details of the algorithms, please see our paper http://arxiv.org/pdf/1305.2532.pdf. 

The code is based on vowpal rabbit library, so you need to install vowpal wabbit. This github code base comes with vowpal wabbit. 

How to install?

First follow the intructions as in the README file to install vowpal wabbit on your computer. 
Then:
cd library
make

You will see the scp executable. Follow the command line and input data format instructions to play with it.