Advanced analytics company QuantumBlack has released its racily-named CausalNex software product.
This is the company’s second open source software development following its previous launch of Kedro, an open source project for data scientists.
So what is CausalNex?
In terms of form and function, CausalNex is a software library designed to allow data scientists and domain experts to co-develop models that go beyond correlation… to consider ‘causal’ relationships.
In the business context, this is argued to help deliver more intelligent interventions that consider the true causes of an organisation’s business challenges.
The company points out that running machine learning projects without considering causality may lead to spurious conclusions, such as stating that more chocolate consumption increases your likelihood to win a Nobel Prize.
CausalNex enables a data scientist to quickly express the dependency between ‘data points’ in a network graph, which can then be inspected and adjusted by a domain expert. This form of hybrid learning with data and domain expertise is meant to ensure models are free from spurious correlations.
Once the structural relationships in the data are established, CausalNex can also use Bayesian Networks to conduct what is known as counterfactual analysis – i.e. asking what happens to target Y if we change feature X.
“One of the biggest benefits of CausalNex is that it generates transparency and trust in models it creates by allowing domain experts to collaborate with data scientist during the modelling process. This helps drive the adoption of recommended interventions. It also avoids making interventions based on spurious correlations. For example, a non-causal model aiming to reduce drought may determine that higher ice cream sales often correspond with higher instances of drought and so spuriously conclude that ice cream sales must be banned,” said the company, in a press statement.
QuantumBlack claims that users especially like the fact that networks can be visualised by common tools, allowing them to understand relationships in their data and work with data scientists to input their expert knowledge as part of the model building process.
CausalNex also streamlines the use of Bayesian Networks for an end-to-end causality and counterfactual analysis, which in the past was a complicated process involving the use of at least three separate open source libraries, each with its own interface.
The company further states that working with Bayesian methods to derive causal inference remains a rich field of industry discourse. The teams says it hopes that open sourcing CausalNex will help contribute to this and ultimately help others enrich their own approach to causality and drive additional value from analytics projects.
CausalNex is a combinatorial play on two words – Causal and Nexus. CausalNex aspires to be the nexus between cause and effect analysis or the nexus of causal reasoning.