User-assisted training data denoising for predictive systems
In one embodiment, a device receives, via a user interface, an indication of what is considered noise within a time series of a path performance metric. The device selects, based on the indication, a particular denoising filter to be applied to telemetry data obtained from one or more network paths regarding the path performance metric. The device forms model training data by applying the particular denoising filter to telemetry data obtained from one or more network paths regarding the path performance metric. The device trains, using the model training data, a prediction model to predict when a given network path will experience a failure condition.
1. A method comprising:
receiving, at a device and via a user interface, an indication of what is considered noise within a time series of a path performance metric;
selecting, by the device and based on the indication, a particular denoising filter to be applied to telemetry data obtained from one or more network paths regarding the path performance metric;
forming, by the device, model training data by applying the particular denoising filter to telemetry data obtained from one or more network paths regarding the path performance metric; and
training, by the device and using the model training data, a prediction model to predict when a given network path will experience a failure condition.
2. The method as in claim 1 , wherein the failure condition of the given network path comprises a service level agreement (SLA) violation or decrease in an application experience metric.
3. The method as in claim 1 , wherein the path performance metric comprises at least one of: jitter, delay, or packet loss.
4. The method as in claim 1 , wherein the indication comprises a selected percentile, and wherein the particular denoising filter is configured to remove measurements of the path performance metric that are below the selected percentile.
5. The method as in claim 1 , wherein receiving the indication of what is considered noise comprises:
providing, by the device, display data to the user interface depicting samples of different denoising filters being applied to the time series, wherein the indication comprises a selection of the particular denoising filter.
6. The method as in claim 1 , wherein the particular denoising filter comprises a wavelet filter or a Savitsky-Golay filter.
7. The method as in claim 1 , further comprising:
determining, prior to training the prediction model, whether the model training data is invalid based in part on quality of experience metrics associated with an online application accessed via the one or more network paths from which the telemetry data was obtained.
8. The method as in claim 1 , wherein forming the model training data comprises:
causing one or more networking devices associated with the one or more network paths to apply the particular denoising filter to the telemetry data obtained from the one or more network paths regarding the path performance metric.
9. The method as in claim 1 , further comprising:
initiating, by device, an adjustment to the particular denoising filter, based on a prediction performance metric for the prediction model.
10. The method as in claim 1 , wherein the prediction model is used to reroute network traffic prior to a predicted failure condition.
11. An apparatus, comprising:
one or more network interfaces;
a processor coupled to the one or more network interfaces and configured to execute one or more processes; and
a memory configured to store a process that is executable by the processor, the process when executed configured to:
receive, via a user interface, an indication of what is considered noise within a time series of a path performance metric;
select, based on the indication, a particular denoising filter to be applied to telemetry data obtained from one or more network paths regarding the path performance metric;
form, by applying the particular denoising filter to telemetry data obtained from one or more network paths regarding the path performance metric, model training data; and
train, using the model training data, a prediction model to predict when a given network path will experience a failure condition.
12. The apparatus as in claim 11 , wherein the failure condition of the given network path comprises a service level agreement (SLA) violation or decrease in an application experience metric.
13. The apparatus as in claim 11 , wherein the path performance metric comprises at least one of: jitter, delay, or packet loss.
14. The apparatus as in claim 11 , wherein the indication comprises a selected percentile, and wherein the particular denoising filter is configured to remove measurements of the path performance metric that are below the selected percentile.
15. The apparatus as in claim 11 , wherein the apparatus receives the indication of what is considered noise by:
providing display data to the user interface depicting samples of different denoising filters being applied to the time series, wherein the indication comprises a selection of the particular denoising filter.
16. The apparatus as in claim 11 , wherein the particular denoising filter comprises a wavelet filter or a Savitsky-Golay filter.
17. The apparatus as in claim 11 , wherein the process when executed is further configured to:
determine, prior to training the prediction model, whether the model training data is invalid based in part on quality of experience metrics associated with an online application accessed via the one or more network paths from which the telemetry data was obtained.
18. The apparatus as in claim 11 , wherein the apparatus forms the model training data by:
causing one or more networking devices associated with the one or more network paths to apply the particular denoising filter to the telemetry data obtained from the one or more network paths regarding the path performance metric.
19. The apparatus as in claim 11 , wherein the process when executed is further configured to:
initiate an adjustment to the particular denoising filter, based on a prediction performance metric for the prediction model.
20. A tangible, non-transitory, computer-readable medium storing program instructions that cause a device to execute a process comprising:
receiving, at the device and via a user interface, an indication of what is considered noise within a time series of a path performance metric;
selecting, by the device and based on the indication, a particular denoising filter to be applied to telemetry data obtained from one or more network paths regarding the path performance metric;
forming, by the device, model training data by applying the particular denoising filter to telemetry data obtained from one or more network paths regarding the path performance metric; and
training, by the device and using the model training data, a prediction model to predict when a given network path will experience a failure condition.