上QQ阅读APP看书,第一时间看更新
Final requirement
Despite the fact that we're model building right now, we want to build with the future in mind. The future is a production-ready machine learning system that performs linear regression. So whatever functions and methods we write have to take into account other things that may occur in a production environment that may not occur in the model -building phase.
The following are things to consider:
- Unseen values: We have to write a function that is able to encode previously unseen values.
- Unseen variables: At some point in the future we might pass a different version of the data in that may contain variables that are unknown at model-building time. We would have to handle that.
- Different imputation strategies: Different variables will require different strategies for guessing missing data.