_pastel

> For one thing, the status of the (so-called) hindsight bias is contested

Why? Were some seminal studies in hindsight bias overturned? This sounds interesting!

...Oh. No, some other guy thought really hard and decided it doesn't make sense. This must be a philosophy paper.

  WITH recursive_traversal(depth, id)
  AS (
    -- base case: the root has no previous
    SELECT 0, id
    FROM todo
    WHERE previous IS NULL
      AND user_id = $1
  
    UNION ALL

    -- recursive case: traverse to next row
    SELECT depth + 1, todo.id
    FROM todo
    JOIN recursive_traversal r ON r.next = todo.id
    WHERE depth < $2
  )
  -- we only accumulated the ids, so join one more time to get the rest of the columns
  SELECT todo.*
  FROM todo
  JOIN recursive_traversal USING (id);

Querying is a bit less straightforward, though; you need a recursive query to traverse the pointers. (Or traverse in application logic, at the cost of a bunch of unnecessary round-trips.)

I've always wanted my own gradient boosting machine implementation to compete with Xgboost and LightGBM. There's a huge space of tricks around regularization, randomization, tree structure, etc., that few people are exploring because neural nets are exploding.

So far I've roughly caught up in speed and accuracy with a few original tricks (and 1/20 of the features), but no real breakthroughs.

Some boards do have a dynamic of "there are roughly X VP we are fighting over; get more than half in the fewest turns". On other boards, if someone were to green too early, they would hopelessly clog up their decks. So players build for as long as possible and gain VP only on the same turn as they can end the game.

There's a wide spectrum between these extremes.

Edit: This game is very relaxing. It goes up to 4 moves at level 90.

Edit 2: I don't think they are autogenerated anymore; level 100 is very aesthetically pleasing. Maybe level 60 was just a mistake or joke?

I've always wondered what proportion of modern real-time video effects rely on ML vs. classical image processing; this not only answers that question, but provides details down to the level of model architecture and the final latency and IOU benchmarks.

Of course I'd be more interested to read how Zoom manages to do even better, but I'm not holding my breath for them to publish those details.

Rather than rejecting outright, did you consider blocking the pull request and asking for benchmarks demonstrating that the optimization was worth the complexity increase? Maybe they would learn more from that process than from an explanation of why they are misguided.

"C's default fall-through in case statements has long been one of its most controversial features; Duff himself said that 'This code forms some sort of argument in that debate, but I'm not sure whether it's for or against.'"

Linear models are simpler. GBMs are more powerful, more flexible, and faster.

Every ML course I took had 3 weeks of problem sets on VC dimension and convex quadratic optimization in Lagrangian dual-space, while decision tree ensembles were lucky to get a mention. Meanwhile GBMs continue to win almost all the competitions where neural nets don't dominate.

I suspect my professors just preferred the nice theoretical motivation and fancy math.

Looks like a rough game. Why did you buy the last Cellar to end the game when you were behind?

If parameters are defined close to usage, and strongly typed, then it's hard to cleanly search for good configurations of the parameters. Especially for fancier search strategies, you want all parameter lookups to go through a single file.

On the other hand, there's a lot of code churn until an ML pipeline is finished. And errors from typos and type violations will often only show up after hours of training. So it's also painful to try to keep a separate, loosely-typed parameter file in sync.

So far, my compromise is to:

(1) on a first pass, define all parameters as global variables at the top of the files they are used in

(2) once mostly code-complete, pull them into a separate file that tracks initial values, current values, and search ranges. Make all usages go through a lookup where the key is an enum, but the value is untyped:

  def param(name: ParamName) -> Any:
    return params[name].current_value

"How different are different 'How different are different diff algorithms in Git' articles?

The caffeinated group outperformed the noncaffeinated on all 5 creativity metrics. The sample size and effect size were too small to prove caffeine caused an improvement.

How did that turn into this title?

At this stage, tactics are mostly about keeping all the rules in your head and thinking hard before each move. But strategy is really interesting. Everyone is guessing blindly, and you can often win by guessing slightly less blindly.

I have two meta-strategies:

(1) Game phases

While learning the rules, whenever possible, mentally categorize game mechanics as opening-related or endgame-related. For example, a lot of board games have some engine-building in the opening and some point maximization in the end.

During the game, constantly estimate the distance to the end of the game. On the first playthrough, most people transition to endgame too late.

(2) Mechanics comparison

Whenever a game has different types of mechanics or resources, search for reference points that compare them. For example, in Dominion, you must choose between buying treasure and action cards. So on the first playthrough, you should deduce that 1 silver is similar to an action with (+2 treasure, +1 action).

The article talks about the metagame transition in Splendor, when players realize that a strategy with minimal engine-building is viable. I deduced that on the first playthrough by trying very hard to estimate the value of a 2-cost card vs taking resources.

Of course, you need to continuously re-evaluate as you understand the game better. By I have a very high winrate on first playthroughs relative to my general ability.

Does anyone else have meta-strategies for this situation?

  def digit_len(n): 
    return len(str(n))

Picnic’s mission is to structure the world’s medical data to make it useful. We work directly with patients to collect, digitize, and manage their complete medical records, giving them with control over their care. Beyond serving patients directly, we partner with biotech, genomics, and pharma companies who sponsor PicnicHealth accounts for research volunteers. Through this work we’re building the data sets that power some of today’s most cutting edge medical research. Our stack is React/Redux, Node, Python, Keras, PostgreSQL, Kubernetes.

Our product has literally saved some of our users' lives by unearthing issues in their medical records. We're growing rapidly — on pace for 20x this year. Our team is smart, hard-working, and passionate about fixing healthcare.

Picnic’s mission is to structure the world’s medical data to make it useful. We work directly with patients to collect, digitize, and manage their complete medical records, giving them with control over their care. Beyond serving patients directly, we partner with biotech, genomics, and pharma companies who sponsor PicnicHealth accounts for research volunteers. Through this work we’re building the data sets that power some of today’s most cutting edge medical research. Our stack is React/Redux, Node, Python, Keras, PostgreSQL, Kubernetes.

Our product has literally saved some of our users' lives by unearthing issues in their medical records. We're growing rapidly — on pace for 20x this year. Our team is smart, hard-working, and passionate about fixing healthcare.

Consider "road incidents". Search is disproportionately higher than deaths - we are definitely interested! But news coverage is extremely low, because what is newsworthy about any specific road incident? It's more naunced than "the news reflects what's interesting".

Or maybe NYT and The Guardian are just missing out on a big area of reader interest, and they could increase engagement by running more stories about road incidents!

Picnic’s mission is to structure the world’s medical data to make it useful. We work directly with patients to collect, digitize, and manage their complete medical records, giving them with control over their care. Beyond serving patients directly, we partner with biotech, genomics, and pharma companies who sponsor PicnicHealth accounts for research volunteers. Through this work we’re building the data sets that power some of today’s most cutting edge medical research. Our stack is React/Redux, Node, Python, Keras, PostgreSQL, Kubernetes.

Our product has literally saved some of our users' lives by unearthing issues in their medical records. We're growing rapidly — on pace for 20x this year. Our team is smart, hard-working, and passionate about fixing healthcare.

Picnic’s mission is to structure the world’s medical data to make it useful. We work directly with patients to collect, digitize, and manage their complete medical records, giving them with control over their care. Beyond serving patients directly, we partner with biotech, genomics, and pharma companies who sponsor PicnicHealth accounts for research volunteers. Through this work we’re building the data sets that power some of today’s most cutting edge medical research. Our stack is React/Redux, Node, Python, Keras, PostgreSQL, Kubernetes.

Our product has literally saved some of our users' lives by unearthing issues in their medical records. We're growing rapidly — on pace for 20x this year. Our team is smart, hard-working, and passionate about fixing healthcare.

Picnic’s mission is to structure the world’s medical data to make it useful. We work directly with patients to collect, digitize, and manage their complete medical records, giving them with control over their care. Beyond serving patients directly, we partner with biotech, genomics, and pharma companies who sponsor PicnicHealth accounts for research volunteers. Through this work we’re building the data sets that power some of today’s most cutting edge medical research. Our stack is React/Redux, Node, Python, Keras, PostgreSQL, Kubernetes.

Our product has literally saved some of our users' lives by unearthing issues in their medical records. We're growing rapidly — on pace for 20x this year. Our team is smart, hard-working, and passionate about fixing healthcare.

Our product has literally saved some of our users' lives by unearthing issues in their medical records. We're growing rapidly — on pace for 20x this year. Our team is smart, hard-working, and passionate about fixing healthcare.