Part 3: Testing
The next step is the feedback and sanity test. A sanity test is a basic test to quickly evaluate whether a claim or the result of a calculation can be valid. It is a simple check to see if my model is rational.
I think that most models available on the open-source platform do not explain the sanity test. ML courses and the ML community mention in passing or simply ignore it altogether. However, any mistake at this stage is crucial because you don’t know what you don’t know. For example, you think that everything looks right, but your final prediction accuracy is very, very low compared to the expectation. You debug your code over and over, and everything seems right only to find missing or Null value in the data. Real-world data are messy, and there are instances where the data is absent. There are many reasons: corrupt data, failure to load the information, or incomplete extraction.