The Violence & Impacts Early Warning System (VIEWS) produces monthly predictions of future violent conflict at both a country and sub-country level of analysis. This repository contains code, configuration files, and documentation that encapsulates the entire process of developing, experimenting, training, evaluating, and deploying the VIEWS machine learning model pipeline.
Use our interactive data dashboard to explore our latest predictions of future armed conflict.
Caution
Please note that this pipeline is actively under construction. We're in the early stages of development, meaning it's not yet ready for operational use. We're working hard to bring you a robust and fully-functional tool, so stay tuned for updates!
- Repository Contents, Structure, and Explanations
- Pipeline Execution
- Pipeline Documentation
- About the VIEWS Project
This repository includes:
- Code: Source code for the VIEWS project's machine learning models and the full pipeline.
- Configuration Files: Settings and configurations for running the models, ensembles, and orchestration scripts.
- Documentation: Detailed instructions and information about the project and how to interact with the pipeline and the individual components.
The VIEWS machine learning pipeline involves several key processes:
- Developing: Creating and refining machine learning models.
- Experimentation: Testing and validating various model configurations and approaches.
- Training: Training models with relevant data.
- Evaluating: Assessing model performance and accuracy.
- Deploying: Implementing models in a production environment to generate monthly true-future forecasts
High-level documentation on the pipeline and its components can be found in the folder documentation
. For a comprehensive understanding of the terms and concepts used, please consult the Glossary
. To explore the rationale behind our architectural choices, visit the Architectural Decision Records (ADRs)
.
Additionally, refer to READMEs and docstrings of various functions and classes in the source code.
The operational fatalities model generates forecasts for state-based armed conflict during each month in a rolling 3-year window. The latest iteration, currently in production, is called Fatalities002.
The following links cover modelling documentation for Fatalities002:
- Prediction models and input variables in main ensemble
- Levels of analysis and dependent variables
- Partitioning and time shifting data for training, calibration, testing/forecasting, model weighting, and out-of-sample evaluation
- Ensembling and calibration
For VIEWS-specific infrastructure documentation, please refer to following GitHub repositories:
ingester3
: Loading input data into the views databaseviewser
: Accessing input data from views databaseviews_api
: Our API for accessing predictions
The VIEWS project is a collaborative effort supported by leading research institutions focused on peace and conflict studies. For more information about the project, visit the VIEWS Forecasting webpage.
-
Peace Research Institute Oslo (PRIO): The Peace Research Institute Oslo (PRIO) conducts research on the conditions for peaceful relations between states, groups, and people. PRIO is dedicated to understanding the processes that lead to violence and those that create sustainable peace. About half of the VIEWS core team is currently located at PRIO.
-
Department of Peace and Conflict Research at the University of Uppsala: The Department of Peace and Conflict Research at the University of Uppsala is a leading academic institution in the study of conflict resolution, peacebuilding, and security. The department is renowned for its research and education programs aimed at fostering a deeper understanding of conflict dynamics and peace processes. This department also hosts the Uppsala Conflict Data Program (UCDP), a central data source for the VIEWS project. About half of the VIEWS core team is currently located at the University of Uppsala.