kyllo

Only perfect multicollinearity (correlation of 1.0 or -1.0) is a problem at the linear algebra level when fitting a statistical model.

But theoretically speaking, in a scientific context, why would you want to fit an explanatory model that includes multiple highly (but not perfectly) correlated independent variables?

It shouldn't be an accident. Usually it's because you've intentionally taken multiple proxy measurements of the same theoretical latent variable and you want to reduce measurement error. So that becomes a part of your measurement and modeling strategy.

In other words, as economists say, because OLS is provably the BLUE (Best Linear Unbiased Estimator) aka the Gauss-Markov Theorem.

You make a good point, though the difference between ML and statistics isn't just about interpreting and validating the model. It's about the "novel discoveries" part aka Doing Science.

Statistical modeling is done primarily in service of scientific discovery--for the purpose of making an inference (population estimate from a sample) or a comparison to test a hypothesis derived from a theoretical causal model of a real-world process before viewing data. The parameters of a model are interpreted because they represent an estimate of a treatment effect of some intervention.

Methods like PCA can be part of that modeling process either way, but analyzing and fitting models to data to mine it for patterns without an a priori hypothesis is not science.

A sailor is sailing her boat across the lake on a windy day. As the wind blows, she counters by turning the rudder in such a way so as to exactly offset the force of the wind. Back and forth she moves the rudder, yet the boat follows a straight line across the lake. A kindhearted yet naive person with no knowledge of wind or boats might look at this woman and say, “Someone get this sailor a new rudder! Hers is broken!” He thinks this because he cannot see any relationship between the movement of the rudder and the direction of the boat.

https://mixtape.scunning.com/01-introduction#do-not-confuse-...

Collider bias or "Berkson's Paradox" is a fun one, there lots of examples of it in everyday life: https://en.wikipedia.org/wiki/Berkson%27s_paradox

Indeed, causally linked variables need not be correlated in observed data; bias in the opposite direction of the causal effect may approximately equal or exceed it in magnitude and "mask" the correlation. Chapter 1 of this popular causal inference book demonstrates this with a few examples: https://mixtape.scunning.com/01-introduction#do-not-confuse-...

This. I bought toothpaste on Amazon, used it up, and the next time I went to buy the exact same toothpaste, the price had doubled.

USPA has a tested division now and it's been gaining in popularity--it will soon be more popular than the untested division if it isn't already. Most of the top untested powerlifters have moved over to the WRPF (which does also have its own tested division). There are a lot of other smaller, regional, untested feds. Then there's the IPF and USAPL and their affiliates, which are fully tested, and are now far more popular than any of the untested feds. Untested might never go away, but tested has rapidly surpassed it in recent years.

Which is based not on your ability to produce value, but your ability to capture value and charge a cut of every unit, and is thus a massive disincentive to produce public goods.

When I lived in South Korea, one of the things that struck me was how much "flatter" the generations there were in terms of pop culture and music taste and awareness, compared to the US. I worked in an office with a bunch of suit-and-tie businessmen who were mostly in their 40s to 60s, and if you were to ask them about any current K-pop group, they all knew their hit songs.

My friend (who's the same age as me) has a 14 year old son who's learning guitar and he asked me for a lesson. The first thing he wanted me to show him was some riffs from AC/DC songs that came out before I was born.

I use plotnine whenever I need to make (static) plots in Python. It's really quite well done, a close match to R's ggplot2, and more feature complete than any of the other Python grammar of graphics packages I've tried.

I'm not sure I've even seen a recent, large C++ project that didn't depend on Python or some other external scripting language just to build it, so it's kind of hard to imagine using C++ itself to solve the problem that using C++ creates.

Ever had a dry-aged steak?

It's a lot easier to go distributed if you're append-only, that's for sure.

Substances affect different people differently. I can have 3-4 drinks in an evening without changing my eating or sleep habits at all and without feeling any different the next day.

Some people experience drinking-related issues other than alcoholism and I just don't want to conflate those issues with alcoholism. I know some actual alcoholics. It's different.

Yeah. When I run out, I just add it to my grocery list and pick up some more the next time I go to the supermarket. Can't remember the last time I left my house specifically to buy alcohol.

A drinking problem isn't defined by quantity, it is when you have a problematic habit/dependence that you can't quit at will. There are plenty of people who enjoy a few drinks several days a week but can just cut back or stop completely when they want/need to because they don't have alcoholism.

I used to drink 3 beers almost every night. I cut down on it to lose weight and now have a single beer 2-3 nights a week and no alcohol other nights. Some weeks I have none. If I had an actual "drinking problem" i.e. alcoholism, I wouldn't have been able to do that.

Except for those people who wake up and have a midnight snack.

Cutting out the snacking is probably also a major factor in the effectiveness of intermittent fasting, especially given that snack foods are often more processed, calorie-dense, and less nutritionally balanced than what we typically eat at meals.

I think a lot of people could lose significant weight by simply eating three meals a day at the usual times, but just cutting out snacking between meals entirely and changing nothing else. Just fasting between meals, if you will.

Power Pivot is the name for the tabular data modeling feature in Excel. It works similarly to Power BI or SQL Server Analysis Services, sharing the underlying database technology. You can load data from queries and files, transform it, add relationships, hierarchies, and calculated measures, and connect pivot tables to it.

Assuming the company has a Microsoft 365 subscription, Mac users can just use the Excel web client in their browser to access the data via pivot tables.

I agree and I do this via Power BI. If you import data into a Power BI report, create a data model with calculated measures (in DAX, not MDX), and publish it to the online service, then users can click on "Analyze in Excel" and it downloads an Excel workbook with a pivot table connected to that data model. I provide this to the PMs for the product I work on and they're able to answer a lot of their questions just by pivoting instead of having to write bespoke SQL.

Darts isn't a specific model, it's a wrapper API for a wide variety of forecasting models, and Prophet is one of them. Other models may or may not outperform Prophet depending on the nature of your specific application and your time series data. You really have to test them to know. And Darts facilitates testing many models on the same data by putting them all behind the same API.

Also, Prophet was developed by a very small number of individuals at Facebook, it's not something they invested massive resources into.

Right, what causal inference really gives you is a tool for specifying assumptions about a model of a data generating process and then estimating a parameter representing the effect size (and the uncertainty surrounding it), provided that your specified assumptions are approximately correct and your measurements are sufficiently accurate.

It never directly answers a "why" or "how" type question. You provide the why/how and then use data to estimate "by how much?"

The Costco near my house sells a large box of frozen mutton cubes, which is the only mutton product I've ever seen at a store in America, and I believe it's mostly purchased by South Asian and Middle Eastern immigrant customers.

I actually think Google picked the winner by deciding to use Python, especially for TensorFlow, right at the time when deep learning really started taking off, and also at a time when they were seen as the leader in engineering best practices that other companies wanted to emulate.

Because it's a power law distribution and because wealth inequality is getting even worse over time.

Are you familiar with the edge popup algorithm introduced in "What's Hidden in a Randomly Weighted Neural Network?" https://arxiv.org/abs/1911.13299v2

Seems relevant to what you're working on. It starts with a randomly initialized, overparameterized neural net, but instead of gradient descent backpropagation, it learns by deleting connection edges.

The data types make sense to statisticians. Ordered factors are great when you need to fit, say, an ordered logit regression model.

A lot of programmers dislike R for the same reason they dislike PHP: weak typing.

If you can get over that, though, there's a lot to like about R. It's sort of like a hybrid between a Lisp and an array language.