Hands-on Machine Learning with Scikit-Learn and TensorFlow

Good read, it’s a classic; it covered all the main points, building up progressively and providing plenty of code. I might reread it since the final TensorFlow section was fairly challenging.

Notes

• Python and statistics basics: fine, clear, especially when it covers MSE and the usual metrics; I skipped the Python parts
• Exploratory analysis was interesting; I learned a few things about matplotlib
• Data cleaning: fine, the usual pandas, while PCA felt tedious and not very useful to me
• Very useful information on encoding, dummy variables, and the importance of standardization
• Regression: the usual topics, though the focus on the math was a nice addition
• SVMs: I have not used them yet
• Decision Trees and Random Forests: very good; it also covered other ensemble models, which seem the most interesting in terms of performance
• Excellent section exploring TensorFlow, especially the math behind functions and a lot of tweaking
• Perceptron and basic concepts: fine; I could follow along
• CNNs were already quite challenging, but the tweaking and the explanation of hyperparameters were excellent
• RNNs: fine, very focused on translation; I believe LSTMs are included
• The autoencoder section was new to me; I’m not sure they are still widely used to generate new material now that there are ANNs
• The finale on reinforcement learning was very interesting but challenging, as it had a lot of code.

Not an easy read, but I will definitely reuse the code if I use TensorFlow in a more advanced way; for now I have not trained anything.