Skip to content

Latest commit

 

History

History
59 lines (45 loc) · 2.34 KB

README.md

File metadata and controls

59 lines (45 loc) · 2.34 KB
Raw

eyetracking-utils

This package is a set of utilities for the eyetracking project. To install, use this command:

!pip install git+https://{user}:{token}@github.com/jspsych/eyetracking-utils.git

Replace {user} with your Github user and {token} with a personal access token that has repo permissions. To access a specific branch, add @{branch_name} to the end of the package.

If you ever make changes to the code, make sure to increment the version in pyproject.toml. Otherwise, the changes will not update when you rerun the pip command.

To import various files, just use:

import et_util.filename as something

And access the functions with:

something.function()

Pipeline Guide

This is how to create TensorFlow datasets using the eyetracking-utils package.

We start from a directory of jpg images uploaded from OSF. The images are the last frame from each webm video recorded during the experiment.

First, make an empty directory where the TFRecord files will be stored. Then, run a function from the tfrecords_processing.py file, such as:

extract_meshes_to_tfrecords(in_path = path/to/jpg/directory, out_path = path/to/empty/directory)

This creates a directory of TFRecord files with one file per subject.

Using functions from the dataset_utils.py file, you can create TensorFlow datasets from the directory of TFRecord files:

train_data, validation_data, test_data = process_tfr_to_tfds(directory_path = path/to/tfrecords, process = parse_tfr_element_mediapipe)

These functions require a process argument, which is a function that parses the TFRecords. The process function must match the tfrecords_processing.py function you chose earlier. You now have 3 TensorFlow datasets: training, validation, and testing data.

Before passing a TensorFlow dataset to a model, you must batch the dataset.

train_data.batch(batch_size)

Versioning

Until we release our v1 of this project, we will use a separate versioning system for this. On development branches, when pushing code, increment the last digit to ensure the code changes will update. Example: version = "0.1.0" -> version = "0.1.1"

Before you merge changes to the main branch, make sure you update the second digit in order to properly update the package for all. Example: version = "0.1.12" -> version = "0.2.0"