The objective function is what EvalML maximizes (or minimizes) as it completes the pipeline search. As it gets feedback from building pipelines, it tunes the hyperparameters to build optimized models. Therefore, it is critical to have an objective function that captures how the model’s predictions will be used in a business setting.
List of Available Objective Functions¶
Most AutoML libraries optimize for generic machine learning objective functions. Frequently, the scores produced by the generic machine learning objective diverge from how the model will be evaluated in the real world.
In EvalML, we can train and optimize the model for a specific problem by optimizing a domain-specific objectives functions or by defining our own custom objective function.
Currently, EvalML has two domain specific objective functions with more being developed. For more information on these objective functions click on the links below.
Build your own objective Functions¶
Often times, the objective function is very specific to the use-case or business problem. To get the right objective to optimize requires thinking through the decisions or actions that will be taken using the model and assigning the cost/benefit to doing that correctly or incorrectly based on known outcomes in the training data.
Once you have determined the objective for your business, you can provide that to EvalML to optimize by defining a custom objective function. Read more here.