Iz članka:

Applying machine learning effectively is tricky. You need data. You need a robust pipeline to support your data flows. And most of all, you need high-quality labels. As a result, most of the time, my first iteration doesn’t involve machine learning at all.

Regardless of whether you’re using simple rules or deep learning, it helps to have a decent understanding of the data. Thus, grab a sample of the data to run some statistics and visualize! (Note: This mainly applies to tabular data. Other data such as images, text, audio, etc. can be tricker to run aggregate statistics on.)

• Simple correlations help with figuring out the relationships between each feature and the target variable.
• Scatter plots are a favorite for visualizing numerical values.
• If either variable is categorical, I’ve found box plots to work well.
• With an understanding of the data, we can then start by solving the problem with heuristics.
• These heuristics also help with bootstrapping labels (aka weak supervision).

When should we use machine learning then? After you have a non-ML baseline that performs reasonably well, and the effort of maintaining and improving that baseline outweighs the effort of building and deploying an ML-based system.