Welcome to daart’s documentation!¶
The daart
package is a collection of tools for the discrete classification of animal behaviors
using low-dimensional representations of videos (such as skeletons provided by tracking algorithms).
daart
combines strong supervision, weak supervision, and self-supervision to improve model
performance.
See the preprint for more details.
This repo currently supports fitting the following types of base models on behavioral time series data:
dense MLP network with initial 1D convolutional layer
RNNs - both LSTMs and GRUs
temporal convolutional networks (TCNs)