4 ms·
I think a great many of astronomers were in fact "data scientists". For example, Kepler's life's work in the end amounts to finding the simplest possible model
by stiff 13y ago
I think a great many of astronomers were in fact "data scientists". For example, Kepler's life's work in the end amounts to finding the simplest possible model that could plausibly account for Tycho de Brahe's data. In fact, many of the most basic statistical modelling methods originated in the process of trying to find explanations for astronomical observations, like least squares:
http://en.wikipedia.org/wiki/Least_squares#History http://en.wikipedia.org/wiki/Least_squares#History
In one sense science is just the outcome of correctly applying the scientific method to a given problem, and in that sense data science is perhaps a meta-science, in that it focuses on the part of the scientific method concerned with making inferences from data, and tries to improve it's effectiveness, using process that is itself scientific to some degree (applying mathematical reasoning to find more reliable ways of inference, introducing additional empirical experiments like cross-validation into the inference process itself for confirmation). I think comparing physics to statistics is a categorical error, and at the same time it's worth keeping in mind that modern physics, biology and chemistry certainly rely very heavily on probability theory and mathematical statistics, so those disciplines always brought up as "properly scientific" are in many respects only rigorous to the extent statistics is.
- kyzyl 13y ago> I think a great many of astronomers were in fact "data scientists" I think the data component is a feature of applied physics in general. Physics problems have had a tendency to be extremely data driven, because when you're dealing with things out in the cosmos, or trying to model nuclear structure for the first time, or shed light on the Higgs mechanism, the data is all you have. You often can't see it with your eye, or a microscope. The effect you're observing only lasts a femtosecond, so you conceive of extremely fancy contraptions that measure the secondary effects en masse (if you'll allow the pun), and you end up with fantastically large data sets. That said, when I look at the scientists I knew/know, including astronomers, their solutions often take a different approach that what I see today in the "data science" blogosphere. There, the approach seems highly centered on scaling various methods to work with huge amounts of data, whereas most of scientists seem to do just the opposite. They go to great lengths--such as building massive integrating antennae the size of football fields--to make sure they collect precisely the correct data, and nothing more. They then take that data and apply the technique that they know must work, if their hypothesis is correct.