In terms of Machine Learning more generally, I want to give special recognition to Jake VanderPlas, an astronomer who dives deep into scikit-learn in some multi-hour Youtube-shared tutorials.
Example: https://youtu.be/L7R4HUQ-eQ0 His excellent keynote at Pycon2017: https://youtu.be/ZyjCqQEUa8o Jake does a super-excellent job of showing off the internal consistency of the scikit-learn API, where you can basically use the same code while just swapping in one classifier or regressor for another. He also speaks the jargon pretty flawlessly, to my ears at least, in terms of what's a feature (label) and what's an observation etc., going into both supervised and unsupervised learning scenarios (scikit-learn handles both). Bravo Jake. Allen Downey has great complementary tutorials which go deeper into the statistical thinking behind these ML models. ThinkBayes is fantastic. It's tempting to just mindlessly throw models at data looking for a best fit, and maybe that's all some underpaid cube farmer has time for, but VanderPlas, along with Downey, wisely counsels against that. Stats more than most is a minefield of pitfalls, such as overfitting. If your aim is authentic research, then mindless model-slinging will quickly come up against its own limitations. That's the message I keep getting from experts in the field. Kirby PS: thanks to Steve Holden, I got to visit the astronomy world up close, the form of the Hubble Space Telescope instrumentation team, eager for Python knowledge. These were already programmers, experts with IDL, but IDL is not the hard currency Python is, in the wider job market. For many reasons, astronomers can't put all their eggs in one basket. The Python ecosystem has been a godsend.
_______________________________________________ Edu-sig mailing list Edu-sig@python.org https://mail.python.org/mailman/listinfo/edu-sig
