Login

Recent searches

No recent searches

Search options

Only available when logged in.
fosstodon.org is one of the many independent Mastodon servers you can use to participate in the fediverse.
Fosstodon is an invite only Mastodon instance that is open to those who are interested in technology; particularly free & open source software. If you wish to join, contact us for an invite.

Administered by:

Server stats:

10K
active users

fosstodon.org: AboutStatusProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.7

#replication

3 posts3 participants0 posts today
Public
Dr Mircea Zloteanu 🌼🐝

#statstab #315 What fraction of repeat experiments will have an effect size within the 95% confidence interval of the first experiment?

Thoughts: CV is a great place to learn. We see the same answer as Cumming: 83%

#replicability #confidenceintervals #replication #metascience #estimation #research #science

stats.stackexchange.com/questi

Cross ValidatedWhat fraction of repeat experiments will have an effect size within the 95% confidence interval of the first experiment?Let's stick to an ideal situation with random sampling, Gaussian populations, equal variances, no P-hacking, etc. Step 1. You run an experiment say comparing two sample means, and compute a 95%
Public *
Mario Angst

So, I managed to set up my first replication repository using Nix (made possible by {rix}).
Some thoughts:
- I used to use Docker in my repositories to facilitate replicating the computing environment. In comparison, Nix feels easier. "Download the folder, unzip, nix-build, nix-shell, Rscript main.R" feels incredibly cool and straightforward.
- {rix} is what made this possible. Otherwise, I would have stayed with Docker, because I would argue learning to write Dockerfiles is objectively easier than Nix commands. @brodriguesco rocks.
- I have serious concerns about the impact of using Nix in terms of scaring people off. Especially because while it is argued Nix works on Windows, working on Windows through WSL is *not* working on Windows for most applied researchers.
- I am not completely satisfied on how this repo turned out in the end and will maybe revisit it at some point. This is mostly because I had a lot of trouble using {reticulate} in a Nix setting (apparently a know issue with $PYTHONPATH) and had to resort to having a setup step downloading miniconda for the replicator, which sort of defeats the purpose of Nix.
- On the other hand, I do not like reticulate anyhow and would usually work in Python directly. Sadly, there is no {rix} for uv ;).
- I am looking forward to my first Docker container using Nix soon ;).
#rstats #OpenScience #replication
fediscience.org/@mario_angst_s

FediScience.orgMario Angst (@mario_angst_sci@fediscience.org)Ok, I have just decided to use nix/ rix in this repository I am currently setting up for reproducing a (published!) analysis, which my co-author conveniently forgot to set up before he left research 😡. Wish me luck. Also, nix is probably the point where no one around me will ever be able to collaborate on stuff with me anymore, since Docker was already a step too far for most. I entirey blame @brodriguesco@fosstodon.org for producing tools that are too appealing not to use. #rstats
Public
Stefano Zacchiroli

Very interesting and important #replication experiment by @huggingface, who is building a #deepseek R1 equivalent #opensource model, with all the needed pieces (yes, including the datasets): huggingface.co/blog/open-r1

(Strictly speaking: it will not be fully end-to-end reproducible, because the distillation part in step 1 starts from Deepseek, which is itself non-reproducible. But it will be *much* better than the status quo, and will produce as byproducts many useful reasoning datasets.)

huggingface.coOpen-R1: a fully open reproduction of DeepSeek-R1We’re on a journey to advance and democratize artificial intelligence through open source and open science.
ExploreLive feeds
About