Complex Data Pipelines
Some data pipelines can be extremely complicated. Data can originate in numerous sources, and go through various phases of preprocessing before being sent to the ML model. This type of pipeline can still be validated by Deepchekcs.
Multi Phase Models
In some data driven systems, ML models appear in various phases of the same pipeline. For example, the pipeline can contain different models which provide certain predictions, and then these predictions can be treated as features which are fed into a different ML model. When this is the case, the pipeline can still be validated by Deepchekcs.
In some cases, an individual ML models is good enough to satisfy the relevant business need. However, in cases that require extremely accurate results, it is fairly common to train multiple models, and aggregate their predictions using voting, stacking, etc. This type of pipeline can still be validated by Deepchekcs.
Model combined with business logic
In many cases, ML models are complemented by determinstic rules which are based on business constraints or business logic. Is isn’t rare to have constraints like this which were hard to embed into the ML model during its training phase. This type of pipeline can still be validated by Deepchekcs.