DEPRECATION WARNING
This code example was intended for use by the legacy Skafos platform and is no longer being maintained. On 05/29/2019, a new version of Skafos was released, streamlining model delivery to the edge.
Sign-up for an account, join our Slack community, and explore some example models to get started.
The following repo contains code for training an activity classifier model on Skafos using the Turi Create framework. The example model was trained on data generated from a watchOS data collection experiment. Given a sequence of motion-sensory readings from an edge device, this model will classify the most likely activity: sitting, standing, or moving.
activity_classifier.ipynb- A Python notebook that trains and saves an activity classifier model to use in your app. Start here.utilities/- a directory that contains helper functions used byactivity_classifier.ipynb.advanced_usage/- a directory that contains additional information about this activity classification model, how to incorporate your own data, and additional example models.requirements.txt- a file describing all required Python dependencies.
- The activity classifier creates a deep learning model, including both convolutional and recurrent layers. It's capable of detecting temporal features in sensor data, lending itself well to the task of activity classification.
- Once trained, you can give the model a sequence of motion-sensory readings generated by a handheld device, and it will classify the current activity as either sitting, standing, or moving.
- The main activity classifier example follows a 3-part blog series that details an example watchOS app: data collection, model creation, and deployment.
Please contact us with questions or feedback! Here are two ways:
Also check out Turi Create's documentation on activity classification basics.