Where to go next?¶
This guide has walked you through the Hello World example of edge sensor detection. Now you should be able to:
Collect and annotate sensor data with the Data Capture Lab to build a high-quality curated dataset.
Use the Analytics Studio’s AutoML engine to build an event detection algorithm suitable for your edge device.
Validate your model is working with the SensiML TestApp in real time.
New Event Exercise¶
To practice everything you have learned, we recommend adding a new event to the Slide Demo project, such as tapping on the top of the sensor. Keep in mind that this project was created to detect continuous events when you select your gesture.
Advanced Model Building¶
For data scientists we recommend looking into more of the advanced features within the SensiML Toolkit. In this guide we discussed the graphical interface using Analytics Studio, however, through Analytics Studio Notebook you can specify the features, transforms, training algorithm, and validation methods that go into your model. You can also specify different sampling, feature selection, and data augmentation techniques. To see how this works see the Advanced Model Building tutorial.
We provide a variety of more advanced projects outside of this guide which you can find on the Data Depot. Here you will find full curated datasets for industrial predictive maintenance, sports wearables, and more. These projects are meant to provide examples to help you get started with your own projects. We are constantly adding new datasets and welcome contributions.
Continuous vs Discrete Events¶
As mentioned in Data Collection Planning, the type of event you are trying to detect changes the way you will label your data. The Getting Started Guide shows you how to label continuous events.
After completing the Getting Started Guide you can read more about the different labeling methods for discrete events at the link: Continuous vs Discrete Events
Using Analytics Studio Notebook you can rehydrate a model and see the code/algorithms in a model that was generated using SensiML AutoML.
Copy the Knowledge Pack UUID from the Explore Model page inside the Model Summary tab
Run the following commands in Analytics Studio Notebook to rehydrate the model
from sensiml import SensiML dsk = SensiML() dsk.project = '<Project Name>' dsk.pipeline = '<Pipeline Name>' kp = dsk.get_knowledgepack('<Knowledge Pack UUID>') dsk.pipeline.rehydrate(kp) #Replace <Project Name>, <Pipeline Name>, and <Knowledge Pack UUID> with your own project parameters
This will print out the model code. From here you can modify the code and re-train your model hyperparameters to improve the accuracy. For more information on the programmatic approach to model building see our Advanced Model Building tutorials for the Analytics Studio.