Remove pupil samples that are physiologically unlikely

The intended use of this method is for removing pupil samples that emerge more quickly than would be physiologically expected. This is accomplished by rejecting samples that exceed a "speed"-based threshold (i.e., median absolute deviation from sample-to-sample). This threshold is computed based on the constant n, which defaults to the value 16.

Usage

detransient(eyeris, n = 16, mad_thresh = NULL)

Arguments

eyeris

An object of class eyeris dervived from load().

n

A constant used to compute the median absolute deviation (MAD) threshold.

mad_thresh

Default NULL. This parameter provides alternative options for handling edge cases where the computed properties here within detransient() $mad\_val$ and $median\_speed$ are very small. For example, if $$mad\_val = 0 \quad \text{and} \quad median\_speed = 1,$$ then, with the default multiplier $n = 16$, $$mad\_thresh = median\_speed + (n \times mad\_val) = 1 + (16 \times 0) = 1.$$ In this situation, any speed $p_i \ge 1$ would be flagged as a transient, which might be overly sensitive. To reduce this sensitivity, two possible adjustments are available:

If $mad\_thresh = 1$, the transient detection criterion is modified from $$p_i \ge mad\_thresh$$ to $$p_i > mad\_thresh .$$
If $mad\_thresh$ is very small, the user may manually adjust the sensitivity by supplying an alternative threshold value here directly via this mad_thresh parameter.

Value

An eyeris object with a new column in timeseries: pupil_raw_{...}_detransient.

Details

This function is automatically called by glassbox() by default. If needed, customize the parameters for detransient by providing a parameter list. Use glassbox(detransient = FALSE) to disable this step as needed.

Users should prefer using glassbox() rather than invoking this function directly unless they have a specific reason to customize the pipeline manually.

Computed properties:

pupil_speed: Compute speed of pupil by approximating the derivative of x (pupil) with respect to y (time) using finite differences.
- Let $x = (x_1, x_2, \dots, x_n)$ and $y = (y_1, y_2, \dots, y_n)$ be two numeric vectors with $n \ge 2$; then, the finite differences are computed as: $$\delta_i = \frac{x_{i+1} - x_i}{y_{i+1} - y_i}, \quad i = 1, 2, \dots, n-1.$$
- This produces an output vector $p = (p_1, p_2, \dots, p_n)$ defined by:
  - For the first element: $$p_1 = |\delta_1|,$$
  - For the last element: $$p_n = |\delta_{n-1}|,$$
  - For the intermediate elements ($i = 2, 3, \dots, n-1$): $$p_i = \max\{|\delta_{i-1}|,\,|\delta_i|\}.$$
median_speed: The median of the computed pupil_speed: $$median\_speed = median(p)$$
mad_val: The median absolute deviation (MAD) of pupil_speed from the median: $$mad\_val = median(|p - median\_speed|)$$
mad_thresh: A threshold computed from the median speed and the MAD, using a constant multiplier $n$ (default value: 16): $$mad\_thresh = median\_speed + (n \times mad\_val)$$

Note

This function is part of the glassbox() preprocessing pipeline and is not intended for direct use in most cases. Provide parameters via detransient = list(...).

Advanced users may call it directly if needed.

Examples

demo_data <- eyelink_asc_demo_dataset()

demo_data |>
  eyeris::glassbox(
    detransient = list(n = 16) # set to FALSE to skip step (not recommended)
  ) |>
  plot(seed = 0)
#> ✔ [  OK  ] - Running eyeris::load_asc()
#> ✔ [  OK  ] - Running eyeris::deblink()
#> ✔ [  OK  ] - Running eyeris::detransient()
#> ✔ [  OK  ] - Running eyeris::interpolate()
#> ✔ [  OK  ] - Running eyeris::lpfilt()

#> ! [ SKIP ] - Skipping eyeris::detrend()
#> ✔ [  OK  ] - Running eyeris::zscore()
#> ! Plotting block 1 from possible blocks: 1