The goal of this project is to determine the behavior of sharks by looking at their accelerometer and gyroscope data. This document is a "getting started" guide for someone who has chosen, or been chosen, to help me with this, including a version of me from the far future who forgot that this existed.
First, find the raw accelerometer data and move it into the repo
directory; put it in a foldr named "Accelerometer". Then run
import_all.m to generate ACCEL.MAT.
The main scripts to look at are closeup.m and turning_svm.m. The
first displays a (possibly labeled) slice of data through a variety
of means; the second tries to use the various feature files
(feature_accel.m, feature_analysis.m, ...) to generate an SVM
predictor.
There's also a script, feature_analysis.m, that performs ANOVA on a
single feature.
SLICES.MAT contains various "data slices" - these are pieces of ACCEL.MAT,
often with an accompanying label file or description. You'll use these
when loading chunks of data via load_accel_slice.m or
load_accel_slice_windowed.m.
Due to space considerations, most of the raw data for this project (barring the labels) is not in this repository. The raw accelerometer data and video data are available from Dan Crear and are backed up on both my laptop and backup hard drives.
The "Accelerometer" folder contains the raw data, and is used by the
code, at least for the purpose of transforming it into ACCEL.MAT via
import_all.m. The "Video" folder contains all video data, and is not
used by the code at this point.