allthing's comments

On: Facebook's Little Red Book

[allthing]

A public benefit corporation is sort of something in between

On: Berkshire halves Apple stake, boosts cash to $277B as it gets 'defensive'

[allthing]

Exactly, this is a rebalancing since Apple because a huge portion of Berkshire's portfolio due to it's growth. They became over 40% of Berkshire's holdings, too much for them to be comfortable with.

[chiefalchemist → allthing]

Right. This is as much about Berkshire's portfolio, investment philosophy and portfolio balance as it is about Apple. Probably more the former than the latter.

On: Ask HN: What are some good resources on Recommender Systems?

[allthing]

One good starting point with code to follow along with is the chapter on Recommender Systems in "Dive in to Deep Learning."

https://d2l.ai/chapter_recommender-systems/index.html

[shreyansh26 → allthing]

That's helpful. Thanks!

On: Ask HN: Recommendations for Good TUI Libraries (Text User Interface)?

[allthing]

I'm not really sure any TUI I've ever seen could really be classified as "good" but the one I like best is probably urwid (http://urwid.org/).

On: Ask HN: Quants and traders: Best OS backtesting platform for trading strategies?

[allthing]

Backtrader is pretty good, but eventually you will find that there really isn't anything out there that will do everything you want it to do.

So imo the best strategy is to roll your own.

[hodder → allthing]

Thanks. I’ll look into backtrader. Eventually I’ll be building my own or stapling onto existing solutions.

On: Ask HN: Technical Programming/Science Podcasts?

[allthing]

A science podcast you can try is Physics Frontiers. It gets pretty technical.

On: Dive into Deep Learning

[allthing]

I don't think they are making up arbitrary rules, I think it's problem solving. Brainstorming alternative solutions that are cost effective and solve the problem is a useful exercise. We shouldn't just blindly use machine learning because it's there.

On: Ask HN: Why are pay-walled articles allowed on Hacker News

[allthing]

I have this trick where I just pay for high quality, well researched news because I'm an adult and I realize that not everything should be free and maybe we should pay for some things.

People on HN like to complain about paying in data to use Google and Facebook and yet also consider it an affront that they have to pay money to journalists.

[Xelbair → allthing]

True, i would agree.

If only paying for high quality news and articles guaranteed me that i won't see ads, well at least non-js ads.

Sadly that won't happen because people who can pay for such news have higher disposable income and are better target for advertisers.

[krageon → allthing]

Which one do you pay for? All of them? How do you determine high quality? How do you reconcile the fact that you need to pay before you can read it, thus not being able to evaluate the quality properly? How do you reconcile saying a large news publication produces quality with the amount of errors that are frequently and egregiously present in articles with any sort of depth?

[tofflos → krageon]

I agree with the opening question but I find latter part of the argument unreasonable. You will encounter many, many services in life that require an initial payment. Pay once and then decide whether you want to keep on paying for more. This applies to restaurants, movies, hotels, taxis, etc. The initial payment for pay-walled articles is usually quite small. If you don't know where to start you can always ask a friend, read a review, or join an online community.

[krageon → tofflos]

I opened and closed with the questions what were to me most relevant, which is my mistake as I should have front-loaded both. Your answer is reasonable, except it glosses over the fact that the quality is bad and this is a pervasive issue.

[cryptozeus → allthing]

Yup this

On: Some Traits of Bismarck (1882)

[allthing]

I don't know if we can equate complex geopolitics with creating a code base. Countries are not computers and people are not deterministic algorithms.

[sorokod → allthing]

Right, goals are different, timescales are different, measures of success are different.

On: A visual proof that neural nets can approximate any function

[allthing]

I believe the universal approximation theorem is for a single hidden layer. When more layers are added arbitrary functions can be approximated.

From section 4.6.2 of Tom Mitchell's Machine Learning book: "Arbitrary functions. Any function can be approximated to arbitrary accuracy by a network with three layers of units (Cybenko 1988)."