ESP  0.1
The Example-based Sensor Predictions (ESP) system tries to bring machine learning to the maker community.
ESP Documentation

The ESP system support the use of machine learning for the analysis of real-time sensor data. ESP examples consist of code that specifies the application of machine learning to a particular application domain. An ESP example consists of multiple parts:

  • Machine learning pipeline for a given application. This is an instance of GestureRecogntionPipeline from the GRT ( specified using usePipeline().
  • An input stream from which to receive live sensor data. This is an IStream instance specified by a call to useInputStream().
  • Zero or one output streams to which to send the predictions generated by the machine learning pipeline. This is an OStream instance specified by a call to useOutputStream().
  • Zero or more calibrators, which allow for the calibration of incoming sensor data based on the analysis of calibration samples collected by the user. Specified by the Calibrator instance passed to useCalibrator().
  • Zero or more tuneable parameters, which allow the user to tune the value of various aspects of the application. Specified by the registerTuneable() functions in the tuneable.h file.
  • Zero or one training samples checkers, to provide the user with feedback on the quality of their training data. Specified by a call to useTrainingSampleChecker().
  • Optional advice to the user on the collection of training data, using useTrainingDataAdvice().
  • Whether or not to use leave-one-out scoring of training samples, specified using useLeaveOneOutScoring().

When compiled against the ESP framework and run, these examples generate a graphical interface that allows the user to visualize live sensor data, collect calibration samples, training, and test data, train the machine learning pipeline, adjust the tuneable parameters, and use the example in a larger interactive project.

For an example, see: user_accelerometer_gestures.cpp