Aslan B.
@motorneurosci.bsky.social
PhD Student @ Donders Institute. Interests: a bit of information processing in the human sensorimotor system, a big share of open reproducible science, a nonsignificant amount of stats, a limited amount of causal inference, and a heavy dose of Olympic WL.
Pr( written in rust | described as "blazingly fast") = 1
November 18, 2024 at 11:39 AM
Pr( written in rust | described as "blazingly fast") = 1
Use case? Probably rare: usually branches will have similar resource requirements. In my case, I have a class of models and parameter sets -- some sets imply a more complicated model that needs more resources. So either branch over all with different resources vs. separate the branches.
November 15, 2024 at 4:33 PM
Use case? Probably rare: usually branches will have similar resource requirements. In my case, I have a class of models and parameter sets -- some sets imply a more complicated model that needs more resources. So either branch over all with different resources vs. separate the branches.
Each of these targets does the same boring thing: call some "expensive function" that uses as many cores as it gets, and log out some details. >
November 15, 2024 at 4:27 PM
Each of these targets does the same boring thing: call some "expensive function" that uses as many cores as it gets, and log out some details. >
I have have two sets of resources: slurm1 allows to submit jobs with 1 CPU, slurm2 allows 2 CPUs. Targets t_a, t_b, t_c are the result of static branching. They all receive 2 CPUs -- there is no way to modify this (?). Targets t_x, t_y, t_z are effectively the same, but only t_x gets 2 CPUs. >
November 15, 2024 at 4:21 PM
I have have two sets of resources: slurm1 allows to submit jobs with 1 CPU, slurm2 allows 2 CPUs. Targets t_a, t_b, t_c are the result of static branching. They all receive 2 CPUs -- there is no way to modify this (?). Targets t_x, t_y, t_z are effectively the same, but only t_x gets 2 CPUs. >
New TIL for {targets} #rstats: For static branching, you'd use tarchetypes::tar_map. But what if you want to assign different resources to those branches? You cannot do this with tar_map (please correct if wrong). Fortunately, you can create you list of targets in any way, including purrr::map... >
November 15, 2024 at 4:17 PM
New TIL for {targets} #rstats: For static branching, you'd use tarchetypes::tar_map. But what if you want to assign different resources to those branches? You cannot do this with tar_map (please correct if wrong). Fortunately, you can create you list of targets in any way, including purrr::map... >
#rstats {targets} tip: when rewriting dynamic branches as static branches via tar_map, use tibble::lst( mapped_targets = tar_map(...), tar_combine(combined_name, tar_select_targets(mapped_targets, ))) to easily combine the branches.
November 10, 2024 at 10:51 PM
#rstats {targets} tip: when rewriting dynamic branches as static branches via tar_map, use tibble::lst( mapped_targets = tar_map(...), tar_combine(combined_name, tar_select_targets(mapped_targets, ))) to easily combine the branches.
Kind of. Intuitively, if you start at random state X and do a certain "random computation" A, then the random state will always change to Y. If you start at Y and do B, the random state will always change to Z. So in your analogy, the sequence exists only *given your sequence of computations*.
November 10, 2024 at 11:27 AM
Kind of. Intuitively, if you start at random state X and do a certain "random computation" A, then the random state will always change to Y. If you start at Y and do B, the random state will always change to Z. So in your analogy, the sequence exists only *given your sequence of computations*.
"set seed then do computations" will yield reproducible results for computations, so:
November 10, 2024 at 10:07 AM
"set seed then do computations" will yield reproducible results for computations, so:
Basically sth like this. But this is just for constructing the 2-sided CI from two 1-sided CIs, IMO this is best mentioned after the actual explanation as a shortcut to assess the test (otherwise it opens more Qs, e.g. how make a 5% inference based on a 10% alpha CI).
November 1, 2024 at 7:26 AM
Basically sth like this. But this is just for constructing the 2-sided CI from two 1-sided CIs, IMO this is best mentioned after the actual explanation as a shortcut to assess the test (otherwise it opens more Qs, e.g. how make a 5% inference based on a 10% alpha CI).
October 30, 2024 at 10:58 AM
Actually, there is a problem with the latest version of quarto on nix (github.com/NixOS/nixpkg...). So, can you pin a specific older version of quarto? Yes, no problem whatsoever.
October 23, 2024 at 7:47 AM
Actually, there is a problem with the latest version of quarto on nix (github.com/NixOS/nixpkg...). So, can you pin a specific older version of quarto? Yes, no problem whatsoever.
And if you don't want to pass cols as a character vector, this would be the version using NSE (I guess, unless there is a more efficient way to write this)
October 12, 2024 at 5:33 PM
And if you don't want to pass cols as a character vector, this would be the version using NSE (I guess, unless there is a more efficient way to write this)
Say you have a dfr with each row a trial, a function f that computes something for one trial (e.g. loglik) but requires additional params that are not in the dfr, an outer function g that applies any such function f to each trial (e.g. to compute total loglik). I think this is the way?
#rstats
#rstats
October 11, 2024 at 2:50 PM
Say you have a dfr with each row a trial, a function f that computes something for one trial (e.g. loglik) but requires additional params that are not in the dfr, an outer function g that applies any such function f to each trial (e.g. to compute total loglik). I think this is the way?
#rstats
#rstats
