Twitter | Search | |
Richard McElreath
Evolutionary Anthropology, behavioral ecology, Bayes book: Occasional blog:
2,369
Tweets
447
Following
5,220
Followers
Tweets
Richard McElreath May 25
I rate these photos magnitude 7 on the squee scale. (I haven't read the paper closely yet, but looks very cool.)
Reply Retweet Like
Richard McElreath retweeted
Juraj Medzihorsky May 25
Had a great time at giving a workshop on Bayesian statistics with based on Statistical Rethinking
Reply Retweet Like
Richard McElreath retweeted
Riana Minocher May 24
Cultural diffs b/w labs make it fun to move around. I definitely miss fika!!!
Reply Retweet Like
Richard McElreath May 24
We worked very hard in this analysis to propagate all the latent-variable uncertainty, and it made a difference by hedging our inferences. makes an appearance. Code and data:
Reply Retweet Like
Richard McElreath May 24
Replying to @marcomarconi83
In this case, I think the range of the data make it hard to recover full function. But might also be some non-identification due to non-additive relationships of parameters.
Reply Retweet Like
Richard McElreath retweeted
Andy Gardner May 24
... and see also the News & Views by -- a very lucid and balanced assessment of what our study shows and what it doesn't.
Reply Retweet Like
Richard McElreath retweeted
F Rodriguez-Sanchez May 24
Very interesting review on stochasticity by From noise to knowledge: how randomness generates novel phenomena and reveals information HT
Reply Retweet Like
Richard McElreath May 24
Visiting my department today is , presenting some of her old and new work on social transmission of adaptive behavior among BaYaka.
Reply Retweet Like
Richard McElreath retweeted
Will Cole May 23
Real life Monty Hall problem. One of these breakfast tacos has steak. I want that one. I pick the far left. Wife tells me far right taco is black bean. I switch to the middle taco. I win steak.
Reply Retweet Like
Richard McElreath May 24
Replying to @dnlmc
In dynamical systems, Pearl's DAGs don't work, because often everything "causes" everything else, and causation can be looped on very short timescales. So abstraction from nature is required to get a system that can be studied with a DAG.
Reply Retweet Like
Richard McElreath May 24
Replying to @dnlmc
I think the Pearl calculus is very useful, but it doesn't represent how causation actually works in nature. Rather, it represents how we symbolically construct causal concepts. The distinction is important. >
Reply Retweet Like
Richard McElreath May 24
Replying to @dnlmc
But I also think Herbert Simon argued effectively that real animals do not reason, in the classical sense nor using Bayesian networks. So I don't see why AI needs to mimic our culturally-constructed, historically bounded vision of rationality. >
Reply Retweet Like
Richard McElreath May 24
Replying to @dnlmc
I haven't read the book yet, and not my area of expertise. As a UCLA grad, I have a lot of sympathy w Pearl's ideas usually. >
Reply Retweet Like
Richard McElreath May 23
Really interesting result of model is that social intelligence can make big brains easily, but has trouble generating energy surpluses. Really promising new approach.
Reply Retweet Like
Richard McElreath retweeted
Mauricio G Forero May 23
A very nice overview of our new paper by :
Reply Retweet Like
Richard McElreath retweeted
Mauricio G Forero May 23
Here's a blog post communities on the behind the scenes of our new brain paper.
Reply Retweet Like
Richard McElreath retweeted
Mauricio G Forero May 23
Did social complexity drive human brain expansion? Contrary to prevailing views, we find evidence against this. New paper with done at funded by and .
Reply Retweet Like
Richard McElreath May 22
Adding parameters can make a model LESS flexible, reduce overfitting. Example from p360 of my book. This came up recently while I was visiting 's lab in Zurich. I still owe you a better motivated example.
Reply Retweet Like
Richard McElreath May 22
You may not care about physics itself, but if you care about science reform, you might care about what is wrong with physics.
Reply Retweet Like
Richard McElreath May 22
Replying to @rtorkar
Not WAIC(...,loglik=TRUE), but rather map2stan(...,log_lik=TRUE). Then you should get log_lik in samples. In general, WAIC and LOO are not trying to estimate same thing. So not surprising that they sometimes differ.
Reply Retweet Like