In hejtmy/fgu-avoidance: Loading, preprocessing and visualising logs from the avoidance box

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(fgu.avoidance)
animals <- load_data(file.path("..", "inst", "extdata", "run", "three-animals.csv"))
animal <- animals$animal_8

Freezes search

Firstly we have to look at the speed the animal makes

plot_speed(animal)

it is evident that the speed kinda jumps around, in some occasions even jumping above 400 units. There are many preprocessing elements which we can do to smooth or fix things up. Please see the package navr for some deeper ideas about speed cleaning and searching for movement stops and onsets. In this vignette, we will only cover the basics

Let's search for all freezings of at least 2s, which don't overcome the speed 0.5. The speed above 0 is NECESSARY, because of how the mechanism actually records animal's position. The device does NOT record consecutively, but only records changes in position. Therefore each two recordings will be from different positions and therefore there will always be some distance traveled and speed above 0.

freezes <- collect_freezes(animal, speed_threshold = 0.5, min_duration = 2)
str(freezes)

So we have found r length(freezes$time) instances of animal freezing. Let's visualise them

plot_speed(animal) +
  geom_navr_timeseries_events(freezes$time_since_start, durations = freezes$duration, color="red", size= 2) +
  xlim(1000,1300)

We can also plot the positions of the animal during those times of

freezes_times <- rbind(freezes$time, freezes$time+freezes$duration)
plot_path(animal) +
  geom_navr_path_events(animal$position, freezes_times)

Speed smoothing

The speed smoothing is optional and unwanted in some situations, but it allows to smooth out some jittery movement in too short times. E.g. if the animal moves 1cm, but because of the recording fault, it moves in 0.001s (say the two recordings were made by erro quite fast one after another), this speed will look like a speed of 10ms/s, although it is just a small step but measured in too short time.

Speed smoothing or position smoothing can solve some of these issues, although it creates potential trouble with smoothing out onsets and potentially removing some freezing period.

The balck line represents the median smoothed value, the blue is the original speed

animal_smooothed <- smooth_speed(animal, type="median", points=7)
plot_speed(animal_smooothed, size=1.25) +
  geom_navr_obj_timeseries(animal$position, "speed", color="blue", alpha = 0.5) +
  xlim(1000,1500) + ylim(0,200)

Because the speeds are smoothed, we need to increase the threshold for the stopping

freezes <- collect_freezes(animal_smooothed, speed_threshold = 3, min_duration = 2)
plot_speed(animal_smooothed) +
  geom_navr_timeseries_events(freezes$time_since_start, durations = freezes$duration, color="red", size= 2)

Speed selection

Speed selection for freeze onset is also quite individual, depending on the current paradigm and especially frequency of recording. If the frequency is quite high, we are runnign into trouble with weird peaks, if too low, then we can have long periods of "nothing".

All in all, the speed needs to be selected per paradigm basis, investigated individually and corrected for potential outliers.

freezes <- collect_freezes(animal, speed_threshold = 0.5, min_duration = 2)
freezes_times <- rbind(freezes$time, freezes$time+freezes$duration)
plot_path(animal) +
  geom_navr_path_events(animal$position, freezes_times)

If we set the speed limit to 5, we will find much more

freezes <- collect_freezes(animal, speed_threshold = 5, min_duration = 2)
freezes_times <- rbind(freezes$time, freezes$time+freezes$duration)
plot_path(animal) +
  geom_navr_path_events(animal$position, freezes_times)

hejtmy/fgu-avoidance documentation built on Sept. 19, 2021, 6:35 a.m.