run_searchlight

Execute a searchlight analysis.

This function runs a searchlight analysis using a specified MVPA model, radius, and method. It can be customized with a combiner function and permutation options.

This function runs a searchlight analysis using a specified RSA model, radius, and method. It can be customized with permutation options, distance computation methods, and regression methods.

This function runs a searchlight analysis using a specified MANOVA model, radius, and method. It can be customized with permutation options.

This function runs a searchlight analysis using a specified vector RSA model, radius, and method. It can be customized with permutation options, distance computation methods, and regression methods.

Usage

run_searchlight(model_spec, radius, method, niter, ...)

# S3 method for mvpa_model
run_searchlight(
  model_spec,
  radius = 8,
  method = c("randomized", "standard"),
  niter = 4,
  combiner = "average",
  permute = FALSE,
  ...
)

# S3 method for rsa_model
run_searchlight(
  model_spec,
  radius = 8,
  method = c("randomized", "standard"),
  niter = 4,
  permute = FALSE,
  distmethod = c("spearman", "pearson"),
  regtype = c("pearson", "spearman", "lm", "rfit"),
  ...
)

# S3 method for manova_model
run_searchlight(
  model_spec,
  radius = 8,
  method = c("randomized", "standard"),
  niter = 4,
  permute = FALSE,
  ...
)

# S3 method for vector_rsa
run_searchlight(
  model_spec,
  radius = 8,
  method = c("randomized", "standard"),
  niter = 4,
  permute = FALSE,
  ...
)

Arguments

model_spec: An object of type vector_rsa_model specifying the vector RSA model to be used.
radius: The radius of the searchlight sphere (default is 8, allowable range: 1-100).
method: The method used for the searchlight analysis ("randomized" or "standard").
niter: The number of iterations for randomized searchlight (default is 4).
...: Additional arguments to be passed to the function.
combiner: A function that combines results into an appropriate output, or one of the following strings: "pool" or "average".
permute: Whether to permute the labels (default is FALSE).
distmethod: The method used to compute distances between searchlight samples ("spearman" or "pearson").
regtype: The method used to fit response and predictor distance matrices ("pearson", "spearman", "lm", or "rfit").

Value

A named list of NeuroVol objects, where each element contains a performance metric (e.g. AUC) at every voxel location.

References

Bjornsdotter, M., Rylander, K., & Wessberg, J. (2011). A Monte Carlo method for locally multivariate brain mapping. Neuroimage, 56(2), 508-516.

Kriegeskorte, N., Goebel, R., & Bandettini, P. (2006). Information-based functional brain mapping. Proceedings of the National academy of Sciences of the United States of America, 103(10), 3863-3868.

Examples

# TODO: Add an example
 
dataset <- gen_sample_dataset(c(4,4,4), 100, blocks=3)
cval <- blocked_cross_validation(dataset$design$block_var)
model <- load_model("sda_notune")
mspec <- mvpa_model(model, dataset$dataset, design=dataset$design, model_type="classification", crossval=cval)
res <- run_searchlight(mspec, radius=8, method="standard")
#> INFO [2024-04-23 13:39:56] model is: sda_notune
#> INFO [2024-04-23 13:39:56] running standard searchlight with 8 radius 
#> INFO [2024-04-23 13:39:56] creating standard searchlight
#> INFO [2024-04-23 13:39:56] running standard searchlight iterator
#> INFO [2024-04-23 13:39:56] mvpa_iterate: compute analysis for batch 1 ...
#> INFO [2024-04-23 13:39:57] mvpa_iterate: compute analysis for batch 2 ...
#> INFO [2024-04-23 13:39:57] mvpa_iterate: compute analysis for batch 3 ...
#> INFO [2024-04-23 13:39:57] mvpa_iterate: compute analysis for batch 4 ...
#> INFO [2024-04-23 13:39:58] mvpa_iterate: compute analysis for batch 5 ...
#> INFO [2024-04-23 13:39:58] mvpa_iterate: compute analysis for batch 6 ...
#> INFO [2024-04-23 13:39:59] mvpa_iterate: compute analysis for batch 7 ...
#> INFO [2024-04-23 13:39:59] mvpa_iterate: compute analysis for batch 8 ...
#> INFO [2024-04-23 13:40:00] mvpa_iterate: compute analysis for batch 9 ...
#> INFO [2024-04-23 13:40:00] mvpa_iterate: compute analysis for batch 10 ...
#> INFO [2024-04-23 13:40:00] mvpa_iterate: compute analysis for batch 11 ...
#> INFO [2024-04-23 13:40:01] mvpa_iterate: compute analysis for batch 12 ...
#> INFO [2024-04-23 13:40:01] mvpa_iterate: compute analysis for batch 13 ...

# A custom "combiner" can be used to post-process the output of the searchlight classifier for special cases.
# In the example below, the supplied "combining function" extracts the predicted probability of the correct class 
# for every voxel and every trial and then stores them in a data.frame.

if (FALSE) { 
custom_combiner <- function(mspec, good, bad) { 
   good %>% pmap(function(result, id, ...) { 
     data.frame(trial=1:length(result$observed), id=id, prob=prob_observed(result)) 
   }) %>% bind_rows()
}

res2 <- run_searchlight(mspec, radius=8, method="standard", combiner=custom_combiner)
}

Usage

Arguments

Value

References

See also

Examples