For novice ML Engineer/ Data Scientist, usually they think that their job is just finished when they just train a “good” model in terms of accuracy, precision, recall etc..
Wrong ! , your job has just started . For stakeholders (Product-owner, business customer or even normal end-user) they will ask you :
Well , why the model is giving us these predictions / recommendtioan?
Easy to ask, hard to answer. But why? For two reasons
i) It is hard to find the root-cause for a specific recommendation / prediction
ii) It is hard to communicate that in a convincing plain-English manner to stakeholders, esp non-tech ones.
Let me give you some-examples of typical situations/ questions to show how-important this topic matter , esp. to industry/business related ML-Models
Learning-to-Rank context
Customer X was expecting results Y to be in the top position in ranking, why he is getting it in third place . This model recommendations don’t make any sense !Stock-price forecast context
Stock s is predicted to have a huge downfall in price by time t. Our domain expert says it has never happened to this stock. Your model is garbage !
Now , you have a flavor of why the problem is important , in Part II I will break the problem in two main sub-problems, giving an overview of how to attack each one
References