prpip 1.2.2

A package for reconstructing pupil size and handling eye-tracker blinks.

prpip

Reconstruct pupil size during blinks in eye-tracking data with a physiologically inspired approach.

Why this approach

This approach offers a scientifically grounded and adaptive method for reconstructing pupil size data during blinks, which is a significant improvement over traditional interpolation techniques. Unlike conventional methods such as linear, spline, or Akima interpolation, which apply a uniform approach to missing data, our model is based on physiological principles of pupil recovery. Specifically, we use a logarithmic recovery model to simulate the natural recovery process of pupil size following occlusions, such as blinks. This model is grounded in pupillometry research that suggests pupil recovery follows an exponential or logarithmic curve, providing a more biologically plausible reconstruction than standard interpolation methods. Furthermore, our method dynamically adapts to the length of the blink. For short blinks (less than 50 ms), the reconstruction uses a polynomial blending approach that ensures smooth transitions between pre- and post-blink data. For longer blinks (greater than 50 ms), the reconstruction applies a logarithmic recovery model to simulate the gradual recovery of the pupil size, with the addition of Gaussian noise proportional to the size difference between pre- and post-blink data. This noise introduces natural variability into the reconstructed data, reflecting the inherent fluctuations in pupil size, which are often absent in traditional models that overly smooth the data. Additionally, the method includes adaptive boundary smoothing, which minimizes artifacts and ensures continuity at the edges of blink intervals. The ability to adjust key parameters, such as the recovery time constant and noise scale, further enhances the flexibility and accuracy of this method. Thus, this approach provides a scientifically sound, biologically informed, and adaptive solution to pupil data reconstruction, making it superior in both realism and accuracy compared to existing interpolation models.

Features

• Automatically detects blink intervals in eye-tracking data.
• Reconstructs pupil size during blinks using:
  • Logarithmic recovery for long blinks (>50 ms).
  • Linear blending for short blinks (<50 ms).
  • Adds stochastic variability to mimic natural pupil fluctuations.
• Processes individual trials or entire datasets.
• Flexible output:
  • Add a new column for reconstructed data.
  • Replace the original pupil size column with reconstructed values.

---

Changelog

See the Changes in versions

Version 0.0.post1

• Initial release of prpip.
• Implemented logarithmic recovery for long blinks and linear blending for short blinks.
• Added stochastic variability to mimic natural pupil fluctuations.
• Supported batch processing of datasets and individual trials.

Version 1.1.0dev1 - Pre-Release

• Enhanced noise scaling for long-blink reconstructions.
• Added advanced parameter customization (tau, noise_scale).
• Improved boundary smoothing for blink transitions.

Version 1.2.1

• Introduced additional output format options.
• Optimized performance for large datasets.

Installation

Install the latest version of prpip from PyPI:

pip install prpip

---

Quick Start

1. Import the Package

from prpip import process_pupil

2. Process an Entire Dataset

import pandas as pd

# Load the dataset
data = pd.read_csv("input.csv")

# Process all trials in the dataset
processed_data = process_pupil(data)

# Save the processed data
processed_data.to_csv("reconstructed.csv", index=False)

3. Process a Specific Trial

# Process only Trial 88
processed_trial = process_pupil(data, trial=88)

# Save the reconstructed trial
processed_trial.to_csv("trial_88_reconstructed.csv", index=False)

4. Plot the Results

import matplotlib.pyplot as plt

plt.figure(figsize=(14, 7))
plt.plot(processed_trial['Timestamp'], processed_trial['Pupil Size'], label='Original Pupil Size', alpha=0.7)
plt.plot(processed_trial['Timestamp'], processed_trial['Reconstructed Pupil Size'], label='Reconstructed Pupil Size', linestyle='--')
plt.xlabel('Timestamp', fontsize=14)
plt.ylabel('Pupil Size', fontsize=14)
plt.title('Original vs Reconstructed Pupil Size (Trial 88)', fontsize=16)
plt.legend(fontsize=12)
plt.grid(True)
plt.show()

---

Input Requirements

The input data must be a Pandas DataFrame or CSV file with the following columns:

• Trial: Identifies the trial number.
• Pupil Size: The measured pupil size.

Output

The output DataFrame includes a new column:

• Reconstructed Pupil Size: Contains the reconstructed values during blinks.

Alternatively, you can replace the original Pupil Size column with the reconstructed values.

---

Advanced Parameters

You can customize reconstruction behavior by adjusting the following optional parameters:

• trial: Specify a trial number to process. If None, all trials are processed.

• blink_threshold: Threshold for detecting blinks. Default is 0 (blinks occur when Pupil Size is 0).

• tau: Recovery time constant for logarithmic reconstruction. Default is 50.

• noise_scale: Scale of Gaussian noise added to long-blink reconstructions. Default is 0.05.

Example:

processed_data = process_pupil(
    data,
    trial=88,
    blink_threshold=0,
    tau=60,
    noise_scale=0.1
)

---

License

This project is licensed under the MIT License.

---

Contributing

We welcome contributions! To contribute:

1. Fork the repository on GitHub.
2. Create a new branch for your feature or bugfix.
3. Submit a pull request.

---

Author

• Mohammad Ahsan Khodami
• Email: [e-mail]
• GitHub: AhsanKhodami

---

Example Input and Output

Input:

Trial	Timestamp	Pupil Size
1	0	4500
1	10	0
1	20	0
1	30	4800

Output:

Trial	Timestamp	Pupil Size	Reconstructed Pupil Size
1	0	4500	4500
1	10	0	4600
1	20	0	4700
1	30	4800	4800