Systems and methods for target device prediction
Systems and methods for training prediction models are illustrated. One embodiment includes a method for training a prediction model in a network. The method includes steps for receiving context data for a portable device in a system, wherein the context data includes localization data that describes a location of the portable device, identifying a predicted stationary device from several stationary devices that is predicted based on the context data using a prediction model, receiving input identifying a target stationary device from the several stationary devices, generating training data based on the predicted stationary device and the received input, updating the prediction model based on the generated training data.
1. A method for training a prediction model in a media playback system comprising a set of one or more processors, the method comprising:
receiving context data for a portable device in the media playback system, wherein the context data comprises localization data that describes a location of the portable device;
identifying a predicted stationary playback device from a plurality of stationary playback devices using a prediction model that takes the context data as input and predicts a predicted stationary playback device as output;
receiving input identifying a target stationary playback device from the plurality of stationary playback devices;
generating training data, wherein the training data comprises the predicted stationary playback device and the identified target stationary playback device; and
updating the prediction model based on the generated training data.
2. The method of claim 1 , wherein the localization data comprises a matrix of probabilities.
3. The method of claim 1 , wherein selecting the predicted stationary playback device comprises:
providing a set of inputs to the prediction model, the set of inputs comprising the received localization data; and
generating probabilities for each stationary playback device of the plurality of stationary playback devices using the prediction model.
4. The method of claim 1 , wherein selecting the predicted stationary playback device further comprises selecting the stationary playback device with a highest probability.
5. The method of claim 4 , wherein selecting the predicted stationary playback device further comprises transmitting a control signal to the predicted stationary playback device.
6. The method of claim 1 , wherein selecting the predicted stationary playback device further comprises:
ranking at least a subset of the plurality of stationary playback devices based on the generated probabilities; and
providing an ordered listing of the plurality of stationary playback devices to the portable device based on the ranking, wherein the portable device presents the ordered listing in a user interface of the portable device.
7. The method of claim 1 , wherein receiving input identifying a target stationary playback device comprises receiving a control command from the target stationary playback device.
8. The method of claim 1 , wherein receiving input identifying a target stationary playback device comprises receiving a selection of the target stationary playback device from the portable device.
9. The method of claim 1 , wherein generating the training data comprises storing the context data as a training data sample and an identification of the target stationary playback device as a label for the training data sample.
10. The method of claim 9 , wherein updating the prediction model comprises:
passing the training data sample through the prediction model to generate probabilities for each stationary playback device of the plurality of stationary playback devices;
computing a loss based on the generated probabilities and the target stationary playback device; and
updating weights of the prediction model based on the computed loss.
11. The method of claim 9 , wherein generating the training data further comprises storing the predicted stationary playback device, wherein updating the prediction model comprises:
computing a loss based on the predicted stationary playback device and the target stationary playback device; and
updating weights of the prediction model based on the computed loss.
12. The method of claim 1 , wherein the context data further comprises at least one of a time of day, day of the week, and a user identity.
13. A method for training a prediction engine in a media playback system comprising a set of one or more processors, the method comprising:
monitoring a media playback system comprising a plurality of playback devices by localizing a portable device in the media playback system;
determining whether to capture training data based on whether a confidence level for the localizing of the portable device exceeds a threshold value; and
upon determining to capture training data:
identifying a true user interaction;
identifying a predicted user interaction from a prediction model; and
generating training data based on the true user interaction and the predicted user interaction.
14. The method of claim 13 , wherein monitoring the system comprises localizing a portable device within the media playback system to determine a probability matrix, wherein the probability matrix indicates likelihoods that the portable device is nearest to each of the plurality of playback devices, wherein identifying the predicted user interaction comprises feeding the probability matrix as an input to the prediction model.
15. The method of claim 13 , further comprising obtaining context data, wherein the context data comprises a set of data from the group consisting of user data, system state data, and localization data.
16. A portable device comprising:
one or more processors; and
data storage having stored therein instructions executable by the one or more processors to cause the portable device to perform a method comprising:
receiving context data for the portable device, wherein the context data comprises localization data that describes a location of the portable device;
identifying a predicted stationary playback device from a plurality of stationary playback devices in a media playback system using a prediction model that takes the context data as input and predicts a predicted stationary playback device as output;
receiving input identifying a target stationary playback device from the plurality of stationary playback devices;
generating training data, wherein the training data comprises the predicted stationary playback device and the identified target stationary playback device; and
updating the prediction model based on the generated training data.
17. The portable device of claim 16 , further comprising a touch-screen display configured to display a graphical user interface.
18. The portable device of claim 17 , wherein the method further comprises updating the graphical user interface displayed on the touch-screen display based on the predicted stationary playback device.
19. The portable device of claim 17 , wherein receiving input identifying the target stationary playback device comprises receiving input identifying the target stationary playback device via the graphical user interface displayed on the touch-screen display.