TensorFlow

Machine Learning with TensorFlow JS

ML Feature has been updated in v3.1.2 more information will follow soon.

supported in v2.1.0^

Machine Learning window Feelix v3.0.0

Collecting Data

The sensor data from the motor can be collected and used to train a Neural Network. The data can be stored for lated used or exported. Play your effects in the main window, add the Microcontroller (bottom left) you want to record from , select the input variables you want to collect below Inputs (top, middle) and hit Record (toolbar on the right). Now the interactions with the motor will be recorded and added to the data set. The data sets can be trimmed into smaller sections using the trim button (toolbar on the right: scissors). Each set is associated with a classifier to label the data for training purposes, the classifier can be changed using the dropdown in the toolbar above the data graph.

Datasets can be saved locally using the buttons in the toolbar (right). They can also be exported and imported as json file. All datasets that are included in the list will be included in the training process.

Initializing and Training a Model

The basic settings of the model can be adjusted to improve the output of the model. All the data listed below files will be used as training data. Select the Open File button to add data sets that have been saved earlier.

Classifying Data

When the model has been trained with the data, the interactions with the motor can be classified using predefined labels. Select Deploy to test the model.

Filters are not working yet. More features will be added to explore with machine learning in haptic and shape changing interfaces.