link-prediction 1.0.8

Predict links between screens of smartphone applications

Link Prediction

Link Prediction is a Python module that provides baseline models to predict links between the screens of smartphone applications.

1. It provides a heuristically constructed set of link data based on the RICO dataset.
2. It provides several heuristic and learning-based link prediction models that can be used using with hierarchical screen data.

This package is part of a master thesis project titled "Predicting Links for Mobile GUI Prototyping" by Christoph A. Johns at German Research Center for Artificial Intelligence (DFKI) and Aalto University. The project is supervised by Michael Barz and Prof. Antti Oulasvirta.

Requirements

Link Prediction is known to work with Python 3.10.3 and above.

Installing

You can install the package using pip with the following command:

pip install link-prediction

You can them import the necessary packages from link_prediction.

Quickstart

Typically, link prediction involves (1) loading some screen data, (2) loading a pre-trained link prediction model and (3) predicting a link (or score) for a potential link from that data.

# Imports
from link_prediction.datasets import load_rico_links
import joblib

# Load some data
X = load_rico_links(download_external_data=True)

# Load a (pre-trained) model
model = "PageContainsLabel"
clf = joblib.load(f"models/{model}Classifier.joblib")

# Predict labels and scores
y_pred = clf.predict(X)
y_score = clf.decision_function(X)

There is also a CLI for simplified use:

$ python3 -m link-prediction --source-screen path/to/view_hierarchy/324.json --source-element 76d99c7 --target-screen path/to/view_hierarchy/339.json --model PageContainsLabel

Keep in mind, however, that using the link-prediction package as shown above requires the original RICO dataset, the RICO_links dataset and the pre-trained models to be present in the project directory.

Contributors

Christoph A. Johns