psolymos / bsims Goto Github PK

Should allow for user provided nest locations. It can be a coercion method that would turn a 2-column matrix or data frame or list into a bsism_population object. Other argument can be a landscape object (use defaults when missing).

Use cases:

• use real territory mapping
• use fine scale habitat info (rasters, polygons) to determine nest locations, assuming that these fine scale differences do not impact behavior and observation layers

Add events accumulation plot for objects returned by get_events and get_detections.

See https://peter.solymos.org/qpad-book/behavior.html#vocalization-events

This would mean to restart counting for each time interval. See qpad-book for implementation (?). Likely we need to isolate counting parts of code to share among transcription options.

Layer functions should have nested inheritance, get_events, get_detections, get_table should work accordingly.

Add new getters for all user action that would otherwise require knowing the internals:

• get_abund: get total N
• get_dens: get avg D=N/A
• get_area: get total A
• get_nests: nest locations

Semi-interactive workflow when comparing different options:

1. use Shiny GUI to explore different options
2. eject settings from GUI
3. import different settings
4. run and replicate simulations under each setting using a wrapper

Wrapper design:

• 1st arg as (1) bsims object (those settings fixed), (2) settings object as ejected by GUI, or (3) individual arguments
• use formals to match inputs with bsims_* args
• return a closure with accessible settings object (object$settings()) and a simulate method (object$run()).

This closure object can be used to be called multiple times for replication like replicate(B, object$run()).

Not quite clear what to store as output (i.e. just table output? -- there is so much more, so maybe this should be left open).

Why is this important?

Because this is how we can make meaningful comparisons.

Default settings with bsims_all should print better:

bi=bsims_all()

r$> bi
bSims wrapper object with settings:
Warning message:
In max(n) : no non-missing arguments to max; returning -Inf

Could use a left drawer to navigate between (independent) apps, fold in the same structure with tabs or steppers for the bSims layers.

When defining home ranges (i.e. movement according to kernel), add an option to restrict overlap (similar to avoiding certain strata).

1. use Voronoi tessellation around nest locations using low-dependency {deldir} pkg https://cran.r-project.org/web/packages/deldir/
2. accept locations only in intersecting polygons using deldir::which.tile

Why is this important?

To improve biological realism for territorial species.

Topics:

• Homogeneous landscape, single realization (take biostat + QPAD book material)
• Stratified landscape, single realization (needs to be developed)
• Multiple realizations (QPAD & MV framework for estimation; needs to be developed)

I wonder whether EDR for vocalizations vs movement related detections would follow a different detection function depending on size of the bird and coloring relative to background.

It is fine to use a mix of CLI/Shiny to make a bsims_all container object and run many realizations under same settings.

• Provide print method for bsims_all containers
• Provide wrapper that interpolates argument values based on a list of values (i.e. single gradient) or over a data frame (a set of value combos)
• Provide option (function) that gives metric of interest (bias etc) that can be calculated over the sets.

If x is provided, it should identify which layers to skip based on its class.

I.e. take class(x)[1L] to identify which layers to skip when assessing formals in .bsims_all.

Response can be positive, and negative response is also more gradual than currently implemented.

Suggest to remove this option (it is really just a filter applied on the events depending on distance to observer).

A hack can be described as extension, i.e. how to further manipulate objects, but it might be too involved for folks without knowledge of the internals (i.e. that events are in a list of tables, and not in a tidy stacked data frame).

Observed might perceive same individual as multiple ones, and multiple individuals as the same, depending on the spatial distinctness (clustering) of detection events.

One such implementation is using ADPclust:

## fit ADP clustering
library(ADPclust)
## get coords
xy <- do.call(rbind, lapply(1:length(x$events), function(i) {
  cbind(x$events[[i]]$x+x$nests$x[i],x$events[[i]]$y+x$nests$y[i])
}))
## get individual id
i <- do.call(c, lapply(1:length(x$events), function(i)
  rep(i, nrow(x$events[[i]]))))
ad <- adpclust(xy, dmethod = "euclidean")
## number of clusters found
ad$nclust
## classification
tab <- table(inds=i, clust=ad$clusters)

We deal with false negatives through detection, but we do not account for false positives.

See https://doi.org/10.1111/1365-2664.13265 for and example