This infographics is inspired by the book Machine learning yearning from Andrew Ng. This book is full of practical advice.
Something that I find interesting is the shift that is occuring between the old machine learning practice and the new machine learning practice. In the old time, data scientists tried as much as possible to reduce the number of features used and the complexity of machine learning models to prevent overfitting. There were a real trade-off between bias and variance. Nowadays, it is common practice to use as much features as possible and to use complex modern architectures. Overfitting is mainly controlled with regularization techniques.
*Note that the error decomposition is an estimation that works quite well.
I will make a post about error analysis. This is a very important topic and I did not have enough room to explain it in details in the infographics.