[D]Deep learning with Python and Keras – course is different from the book!!
I did a course on Udemy called ‘deep learning with python and keras’ thinking it’s a course version of the similar name book by François Chollet (author of Keras). But it’s not!
It was stupid of me to not realize this earlier. The course is made by some 2 other guys. Having finished the course I think, it’s average at best.
- The worst part is that the examples they use in the course hardly prove the theory. For example, they make a claim on the influence of batch size on accuracy and rate of learning. But when you run their example it does not prove their point.
- Some of the data they use is low in volume and quality that minor changes in test/train split or hyperparameter changes gives huge improvements.
- Another strange thing I noticed was that the test error would be lower than training error in a lot of cases.
Overall I would strongly urge people to learn theory from many other strong courses (like Andrew NG’s) and for Keras, buy the book by the author.
This course was made by two guys, both of who have no experience but teaching experience in ML through these online courses and bootcamps. Neither have academic background not industry experience in ML. Not to say that academic background or industry experience is necessary for being an expert in ML, but if someone doesn’t have either, I’d feel a little skeptical.