Based on your data structure and requirements, I would suggest a solution that uses the imputeTS package in R, which provides an efficient way to handle time series interpolation.

Here’s an example code snippet:

library(imputeTS)

# Identify blink onset and offset
onset <- which(df$BLINK_IDENTIFICATION == "Blink Onset")[1]
offset <- which(df$BLINK_IDENTIFICATION == "Blink Offset")[1]

# Interpolate Pupil_Avg values before blink onset to after blink offset using linear interpolation
df$Pupil_Avg[onset:offset] <- na.interpolation(df$Pupil_Avg, option = "linear")

# Replace -1 values in Pupil_Avg column with NA
df$Pupil_Avg[df$Pupil_Avg == -1] <- NA

# Run imputeTS function to perform interpolation and fill missing values
df <- imputeTS(df$Pupil_Avg, option = "linear")

This code snippet assumes that you have a single blink onset and offset in your time series. If you have multiple instances of blink onset and offset, you can modify the which function calls to extract all occurrences.

Note that I’ve used the imputeTS package instead of the zoo package, as suggested by the original answer. The imputeTS package provides a more efficient way to perform interpolation and fill missing values in time series data.

Also, I’ve added the option = "linear" argument to the na.interpolation function to specify the type of interpolation. You can change this to "spline" or "stine" to use Spline or Stineman interpolation instead.

Finally, I’ve run the imputeTS function on the interpolated data to fill any remaining missing values and perform additional smoothing.

This solution should provide a more efficient and accurate way to interpolate your time series data, especially for large datasets like yours.

Last modified on 2024-07-09