In this page you can find the experiments performed as part of the paper "Training Binary Classifiers as Data Structure Invariants" published at ICSE 2019.

Getting started

Download the tool from the following link: dsclassifier

Prerequisites

Install the following tools:

• Java 7 or greater
• Python 3
• Scikit-learn python package (we recommend to install it using pip)
• Pandas python package (we recommend to install it using pip)

Running the experiments

The three different experiments that we have performed are described in the following pages:

1. Experiment1: Building negatives instances
2. Experiment2: Measuring the neural network performance
3. Experiment3: Improving bug finding

Related work

Other mechanisms to infer specifications:

1. Daikon is an implementation of dynamic detection of likely invariants. Here you can find a comparison of our approach with the Daikon approach.