There are so many things that can go wrong in machine learning project. For example:
- the core of any ML project is an input data pipeline (the input ETL). There can be some row missing in one table and the whole solution will break
- the ML models are computing predictions on new data - we can only assume that they will work as expected, but the true is that no-one knows how they will behave
- the ML models often tend to be very complex = hard to understand = hard to debug (there are some explainers for ML/AI, but it is one more complex module in the project :))
I agree 100% with article's author that you should start with something super-simple that works, and then try to iterate, one thing at a time
- the core of any ML project is an input data pipeline (the input ETL). There can be some row missing in one table and the whole solution will break
- the ML models are computing predictions on new data - we can only assume that they will work as expected, but the true is that no-one knows how they will behave
- the ML models often tend to be very complex = hard to understand = hard to debug (there are some explainers for ML/AI, but it is one more complex module in the project :))
I agree 100% with article's author that you should start with something super-simple that works, and then try to iterate, one thing at a time