Finding candidate classes

Last updated: 2022-07-30

Checks: 4 2

Knit directory: workflowr/

This reproducible R Markdown analysis was created with workflowr (version 1.7.0). The Checks tab describes the reproducibility checks that were applied when the results were created. The Past versions tab lists the development history.

Environment: empty

Great job! The global environment was empty. Objects defined in the global environment can affect the analysis in your R Markdown file in unknown ways. For reproduciblity it’s best to always run the code in an empty environment.

Seed: set.seed(20190717)

The command set.seed(20190717) was run prior to running the code in the R Markdown file. Setting a seed ensures that any results that rely on randomness, e.g. subsampling or permutations, are reproducible.

Session information: unavailable

Recording the operating system, R version, and package versions is critical for reproducibility. To record the session information, add sessioninfo: “sessionInfo()” to _workflowr.yml. Alternatively, you could use devtools::session_info() or sessioninfo::session_info(). Lastly, you can manually add a code chunk to this file to run any one of these commands and then disable to automatic insertion by changing the workflowr setting to sessioninfo: ““.

Cache: none

Nice! There were no cached chunks for this analysis, so you can be confident that you successfully produced the results during this run.

File paths: relative

Great job! Using relative paths to the files within your workflowr project makes it easier to run your code on other machines.

Repository version: no version control

Tracking code development and connecting the code version to the results is critical for reproducibility. To start using Git, open the Terminal and type git init in your project directory.

This project is not being versioned with Git. To obtain the full reproducibility benefits of using workflowr, please see ?wflow_start.

Finding candidate latent classes: the graph enumeration and pruning algorithm

—Special considerations: this portion is likely a high-memory task!—

Now that we have our pairwise fits, we need to figure out what are the remaining candidate latent classes. These can be found using the path enumeration and pruning algorithm. This is a graph-based algorithm which can require a lot of memory (especially when the dimension of the data is large). However, with the memory available, this algorithm runs very quickly – in our experience, in under 5 minutes.

First, load in the previously obtained pairwise fits.

data("fits")

Then, we can obtain a reduced list of candidate latent classes with get_reduced_classes(). This function has 3 arguments:

the pairwise fits
the dimension of the data
the name of an output “LEMON graph format” file (here, called lgf.txt). This is not to be edited by the user, but is produced for underlying C software.

# This finds the dimension of the data directly from the pairwise fits
D <- as.numeric(strsplit(tail(names(fits),1), "_")[[1]][2])

# Get the list of candidate latent classes
red_class <- get_reduced_classes(fits, D, "output/lgf.txt", split_in_two = FALSE)

Writing LGF file...done!
Finding latent classes...done!

# write the output to a text file
readr::write_tsv(data.frame(red_class), file = "output/red_class.txt", col_names = FALSE)

Each row of red_class corresponds to a candidate latent class across the 3 dimensions. The remaining candidate latent classes are as follows:

red_class

      [,1] [,2] [,3]
 [1,]    1    0    1
 [2,]    1    0    0
 [3,]    1    0   -1
 [4,]    0    1   -1
 [5,]    0    0    1
 [6,]    0    0    0
 [7,]    0    0   -1
 [8,]    0   -1    1
 [9,]    0   -1    0
[10,]   -1    0    1
[11,]   -1    0    0
[12,]   -1   -1    1
[13,]   -1   -1    0

which is a subset of the $3^D=27$ candidate latent classes. Next, we need to determine the hyperparameters on the priors in our Bayesian Gaussian mixture model. Most important is computing the hyperparameters for the prior on the mixing weights. Some classes (especially when the dimension is larger) will have a mixing weight that will result in a degenerate mixing distribution, and so classes with small enough mixing weights can be further pruned from the model. This is discussed in the next step.