reviewed in July, 2019 - shortly after the initial (full) release
Machine Learning Yearning is a 58 chapter collection of technical strategies for deep learning projects by Andrew Ng. It is part of deeplearning.ai
The book certainly aims at people without a lot of AI/ML experience. IMHO the target audience of most chapters is more on the management and software engineering side not on the hard AI/ML research/engineering side. It certainly aims to avoid common mistakes, which is something I can’t emphasize enough.
It contains chapters on:
- Introduction to ML/AI problems
- Setting up development and test sets
- Basic Error Analysis
- Bias and Variance
- Learning curves
- Comparing to human-level performance
- Training and testing on different distributions
- Debugging inference algorithms
- End-to-end deep learning
- Error analysis by parts
- Conclusion
It certainly contains a lot of common knowledge and some tips and tricks - some of which are outdated already (deep learning moves fast).
What is doesn’t cover is how to think. Well, from experience, it is really challenging to teach that. IMHO, AI/ML is mainly applied to toy problems at the moment. And yes, I refer to production use. Except for a few cases, I would consider most applications still similar to toy problems. Datasets are messy and there are more restrictions and incompleteness than in Kaggle competitions or online courses but basically it is all very similar.
Since it is free: happy reading and please come up with your own thoughts on topics covered in that book.