3 ms·

A Cookbook for Machine Learning: Vol 1

193 points by forgotmysn 9y ago

meow_mix 9y ago
Idk about you guys but this is not what I think of when I hear the word "cookboook"
- samhunta 9y ago
  I would love to learn more about machine learning as a web developer, but simply put it's a different industry with entirely different constructs. I found a great article[1] that at least helped me build a simplistic understanding as a modern developer. [1]. https://simpleprogrammer.com/2017/10/02/machine-learning-modern-web-developer/ https://simpleprogrammer.com/2017/10/02/machine-learning-mod...
- rrmm 9y ago
  It depends on what you're cooking. All that a cookbook really implies is a template recipe to apply when you are faced with a certain situation. But yeah...It's definitely not a beginner ML cookbook.
glial 9y ago
I really appreciate his sharing his notes that translate fairly complex math into something like an intuition. It fits in a helpful (but sparsely populated!) niche between academic papers and tensorflow-for-software-engineers tutorials.
hasa 9y ago
I must admit I did not understand even the basics of this posting. I Need to start way further behind this level of abstraction.
- keont 9y ago
  Yep. Second formulae and I'm out of my league. Is it just me that gets frustrated when things aren't explained? What do I have to study to be able to read this θ∗,ψ∗←argminθ,ψg(θ,ψ) With no prior explanation and say, yep, now that you put it that way, it's true?
  - pedrosorio 9y ago
    Notation is notation. I think it's hard to come up with a single guide because a lot of conventions depend on the area. In the context of optimization, a star refers to the value of the variable that minimizes (or maximizes) a function you're interested in. You will also see a hat over a variable (https://en.wikipedia.org/wiki/Hat_operator#Estimated_value https://en.wikipedia.org/wiki/Hat_operator#Estimated_value) sometimes (from statistics) which is an estimate of some quantity from data (and that estimate may be obtained by minimizing a function - solving an optimization problem) By the way, this is the first result when googling argmin: https://en.wikipedia.org/wiki/Arg_max https://en.wikipedia.org/wiki/Arg_max
  - zardo 9y ago
    This isn't a formula that is 'true', it's a minimally specified program. It takes a function g, and returns the values of it's arguments (theta and psi) that minimize the function g.
  - master_yoda_1 9y ago
    This might help https://en.wikipedia.org/wiki/List_of_mathematical_symbols https://en.wikipedia.org/wiki/List_of_mathematical_symbols
alexcnwy 9y ago
I studied post-grad ML and I can say you really don't need to go this deep to use ML - check out http://fast.ai http://fast.ai for a code-first practical crash course :)
- backpropaganda 9y ago
  The author fails to mention this, but this is meant to be a cookbook for machine learning research.
- j_s 9y ago
  The discussion 3 months ago on "Part 2" mentioned a re-launch on the "Part 1" content ("redoing the whole of part 1 starting in October"); how's that working out? I wish the tedious maintenance (keeping instructions updated with the latest versions of the software, etc.) was a higher priority there. https://news.ycombinator.com/item?id=14887414 https://news.ycombinator.com/item?id=14887414 Apparently there's been a bit of "broken links and poorly organized content". If you decide to try it out, don't miss http://wiki.fast.ai/ http://wiki.fast.ai/ PS. Ask HN: Best way to get started with AI? | https://news.ycombinator.com/item?id=15689399 https://news.ycombinator.com/item?id=15689399 (3 days ago)
wjnc 9y ago
As someone more at home in statistics than ML/CS, it's funny that I find this easier to understand than the more 'true' cookbook style postings. The real 'grok' for me would be this kind of mathematical intuition, combined with real data examples. I always learn by just trying to recreate models in books, down to the algebra. If you can feel it and inspect it, you'll get it.

193 points by forgotmysn 9y ago

meow_mix 9y ago

Idk about you guys but this is not what I think of when I hear the word "cookboook"

samhunta 9y ago
I would love to learn more about machine learning as a web developer, but simply put it's a different industry with entirely different constructs. I found a great article[1] that at least helped me build a simplistic understanding as a modern developer. [1]. https://simpleprogrammer.com/2017/10/02/machine-learning-modern-web-developer/ https://simpleprogrammer.com/2017/10/02/machine-learning-mod...
rrmm 9y ago
It depends on what you're cooking. All that a cookbook really implies is a template recipe to apply when you are faced with a certain situation. But yeah...It's definitely not a beginner ML cookbook.

glial 9y ago

I really appreciate his sharing his notes that translate fairly complex math into something like an intuition. It fits in a helpful (but sparsely populated!) niche between academic papers and tensorflow-for-software-engineers tutorials.

hasa 9y ago

I must admit I did not understand even the basics of this posting. I Need to start way further behind this level of abstraction.

keont 9y ago
Yep. Second formulae and I'm out of my league. Is it just me that gets frustrated when things aren't explained? What do I have to study to be able to read this θ∗,ψ∗←argminθ,ψg(θ,ψ) With no prior explanation and say, yep, now that you put it that way, it's true?
- pedrosorio 9y ago
  Notation is notation. I think it's hard to come up with a single guide because a lot of conventions depend on the area. In the context of optimization, a star refers to the value of the variable that minimizes (or maximizes) a function you're interested in. You will also see a hat over a variable (https://en.wikipedia.org/wiki/Hat_operator#Estimated_value https://en.wikipedia.org/wiki/Hat_operator#Estimated_value) sometimes (from statistics) which is an estimate of some quantity from data (and that estimate may be obtained by minimizing a function - solving an optimization problem) By the way, this is the first result when googling argmin: https://en.wikipedia.org/wiki/Arg_max https://en.wikipedia.org/wiki/Arg_max
- zardo 9y ago
  This isn't a formula that is 'true', it's a minimally specified program. It takes a function g, and returns the values of it's arguments (theta and psi) that minimize the function g.
- master_yoda_1 9y ago
  This might help https://en.wikipedia.org/wiki/List_of_mathematical_symbols https://en.wikipedia.org/wiki/List_of_mathematical_symbols

alexcnwy 9y ago

I studied post-grad ML and I can say you really don't need to go this deep to use ML - check out http://fast.ai http://fast.ai for a code-first practical crash course :)

backpropaganda 9y ago
The author fails to mention this, but this is meant to be a cookbook for machine learning research.
j_s 9y ago
The discussion 3 months ago on "Part 2" mentioned a re-launch on the "Part 1" content ("redoing the whole of part 1 starting in October"); how's that working out? I wish the tedious maintenance (keeping instructions updated with the latest versions of the software, etc.) was a higher priority there. https://news.ycombinator.com/item?id=14887414 https://news.ycombinator.com/item?id=14887414 Apparently there's been a bit of "broken links and poorly organized content". If you decide to try it out, don't miss http://wiki.fast.ai/ http://wiki.fast.ai/ PS. Ask HN: Best way to get started with AI? | https://news.ycombinator.com/item?id=15689399 https://news.ycombinator.com/item?id=15689399 (3 days ago)

wjnc 9y ago

As someone more at home in statistics than ML/CS, it's funny that I find this easier to understand than the more 'true' cookbook style postings. The real 'grok' for me would be this kind of mathematical intuition, combined with real data examples. I always learn by just trying to recreate models in books, down to the algebra. If you can feel it and inspect it, you'll get it.