Data scientists love doing experiments, training models, and making their hands dirty with data. At the beginning of a project, enthusiasm is at the top, but when things become complicated or too time-consuming, looking for simpler solutions is a real must.
There may be situations where business stakeholders ask to make changes to the underlying solution logic or to make further adjustments/trials while trying to improve performance and maintain a good explicative level of the predictive algorithms involved. Identifying possible bottlenecks in the code implementation, which may lead to additional complexity and delays in delivering the final product, is crucial.
Imagine being a data scientist and having the task of developing a predictive model. We have all that we need easily at our disposal and after a while, we are ready to present to the business people our fancy predictive solutions built on thousands of features and millions of records that achieve astonishing performances.
The business stakeholders are fascinated by our presentation and understand the technology’s potential, but they added a request. They want to know how the model takes its decisions. Nothing easier we may think…