4 ms·

For ML and Data Science I'd say very immature. HLearn does interesting things but is far from providing a workable ML or Data Science workflow. We need solid, u

by tel 11y ago

For ML and Data Science I'd say very immature. HLearn does interesting things but is far from providing a workable ML or Data Science workflow. We need solid, unquestionable Matrix and Vector libs upon which the ~50 bread and butter algorithms cans be solidly implemented. Then Diagrams needs super tight integration and connections to "big data" systems can be built. Frames is moving in the right direction type-wise but it'd be really nice to get HLists (or even vinyl?) in as representations which can be transparently backed by the aforementioned common matrix libs. Lenses could probably help a lot here and I think Anthony is going in that direction exactly. Neatly, the GPU story is actually quite strong and could be integrated with this hypothetical ML and Data Science platform I'm describing quite nicely. That'd be a not small boost to its power.

jordigh 11y ago
What's the difference between ML and data science? ML has more programming? And what's the difference between data science and statistics? Statisticians try to come up with new mathematics, while data scientists just try to come up with ways to apply that maths?
- tel 11y ago
  I'll take a brief stab since I think there are important differences but this is also holy war territory so YMMV. ML tends to be more concerned with scale and computational efficiency as the algorithms used are often (a) heavyweight and (b) less human interpretable. Success is primarily measured objectively but via proxy measures like prediction accuracy. It may require data science efforts behind the scenes in order to succeed. Data Science tends to be more concerned with interpretation than scale or repetition. It sits somewhere between analyst, sataistician, and journalist. It's also a much more interactive workflow often involving a great deal or successive, simple models and lots of exploratory visualization. Measures of success include discovery of human-relatable inferences from data, substantiation or dismissal of human-relevant experimental hypotheses, and success in communication of those ideas. Statistics is a field of study of some of the most important mathematical foundations of each of these roles. A statisicians skill is more generalized but also more abstract. A statistician who learns CS and journalism could easily head into either other field.

by tel 11y ago

jordigh 11y ago

What's the difference between ML and data science? ML has more programming? And what's the difference between data science and statistics? Statisticians try to come up with new mathematics, while data scientists just try to come up with ways to apply that maths?

tel 11y ago
I'll take a brief stab since I think there are important differences but this is also holy war territory so YMMV. ML tends to be more concerned with scale and computational efficiency as the algorithms used are often (a) heavyweight and (b) less human interpretable. Success is primarily measured objectively but via proxy measures like prediction accuracy. It may require data science efforts behind the scenes in order to succeed. Data Science tends to be more concerned with interpretation than scale or repetition. It sits somewhere between analyst, sataistician, and journalist. It's also a much more interactive workflow often involving a great deal or successive, simple models and lots of exploratory visualization. Measures of success include discovery of human-relatable inferences from data, substantiation or dismissal of human-relevant experimental hypotheses, and success in communication of those ideas. Statistics is a field of study of some of the most important mathematical foundations of each of these roles. A statisicians skill is more generalized but also more abstract. A statistician who learns CS and journalism could easily head into either other field.