What ML stack do tech startups use, you ask? Let's look at the Hacker News jobs board.
Out of 964 job postings:
- 12 posts mention TensorFlow
- 7 Keras
- 5 Scikit-Learn
- 1 Caffe
- 0 PyTorch, MXNet
Also note that:
- 89 posts mention ML (9%)
- 34 AI
- 18 deep learning
Regarding #MachineLearning and interpretability: It's easy enough to get relative feature importances from a Random Forest. Are there any promising strategies for finding the *directionality* of a feature in a RF and how it relates to the outcome?