Temporal-clustering invariance in irregular time series data

1. A computer-implemented method comprising:

obtaining, by one or more electronic devices, an input irregular time series having a plurality of data points, wherein each data point of the plurality of data points includes a timestamp and a feature vector;

generating, by the one or more electronic devices, a plurality of variant time series from the input irregular time series, wherein a first variant time series of the plurality of variant time series includes a first data point that is based in part on a combination of at least two data points of the input irregular time series;

generating, by the one or more electronic devices, a first output from a neural network model based on a first variant time series of the plurality of variant time series;

generating, by the one or more electronic devices, a second output from the neural network model based on a second variant time series of the plurality of variant time series; and

outputting, by the one or more electronic devices, a classification vector based in part on the first output and the second output.

2. The computer-implemented method of claim 1 , wherein the first output is scaled by a first weight and the second output is scaled by a second weight.

3. The computer-implemented method of claim 1 , wherein the first data point of the first variant time series includes an indication of a number of data points of the input irregular time series that were combined to form the first data point.

4. A computer-implemented method comprising:

obtaining, by one or more electronic devices, an input irregular time series having a plurality of data points wherein each data point of the plurality of data points includes a timestamp and a feature vector;

generating, by the one or more electronic devices, a plurality of variant time series from the input irregular time series, wherein a first variant time series of the plurality of variant time series includes a first data point that is based in part on a combination of at least two data points of the input irregular time series; and

using, by the one or more electronic devices, the plurality of variant time series for a machine learning task.

5. The computer-implemented method of claim 4 , wherein the machine learning task trains a neural network model, the method further comprising:

for each variant time series of the plurality of variant time series, processing the variant time series with the neural network model to generate an output; and

updating a parameter of the neural network model based at least in part on the output.

6. The computer-implemented method of claim 4 , wherein the machine learning task generates a prediction vector, the method further comprising:

generating a first output from a neural network model based on a first variant time series of the plurality of variant time series;

generating a second output from the neural network model based on a second variant time series of the plurality of variant time series; and

outputting the prediction vector based in part on the first output and the second output.

7. The computer-implemented method of claim 6 , wherein the first output is scaled by a first weight and the second output is scaled by a second weight.

8. The computer-implemented method of claim 4 , wherein generating the first variant time series comprises:

grouping the data points in the input irregular time series into a plurality of clusters, wherein a number of clusters in the plurality of clusters is less than a number of data points in the input irregular time series, wherein each cluster corresponds to a data point in the first variant time series of the plurality of variant time series;

generating, for each cluster, a time stamp from the time stamps of the data points grouped into the cluster; and

generating, for each cluster, a feature vector from the feature vectors of the data points grouped into the cluster.

9. The computer-implemented method of claim 4 , wherein generating the first variant time series comprises:

selecting a plurality of edges in the input irregular time series, an edge corresponding to a time between consecutive data points in the input irregular time series;

generating, for each selected edge, a time stamp from the time stamps of the consecutive data points bounding the selected edge; and

generating, for each selected edge, a feature vector from the feature vectors of the consecutive data points bounding the selected edge.

10. The computer-implemented method of claim 4 , wherein generating the first variant time series comprises:

generating a plurality of clusters, each cluster corresponding to a regularly spaced timestamp between an earliest timestamp of a second data point in the input irregular time series and a latest timestamp of a third data point in the input irregular time series, wherein a number of clusters in the plurality of clusters is less than a number of data points in the input irregular time series;

associating each data point in the input irregular time series with one of the plurality of clusters;

generating, for each cluster, a time stamp from the time stamps of the data points associated with the cluster; and

generating, for each cluster, a feature vector from the feature vectors of the data points associated with the cluster.

11. The computer-implemented method of claim 4 , wherein the first data point of the first variant time series includes an indication of a number of data points of the input irregular time series that were combined to form the first data point.

12. The computer-implemented method of claim 4 , wherein the input irregular time series is one of the plurality of variant time series.

13. A system comprising:

a storage service implemented by a first one or more electronic devices; and

a machine learning service implemented by a second one or more electronic devices, the machine learning service including instructions that upon execution cause the machine learning service to:

obtain an input irregular time series having a plurality of data points from the storage service wherein each data point of the plurality of data points includes a timestamp and a feature vector;

generate a plurality of variant time series from the input irregular time series, wherein a first variant time series of the plurality of variant time series includes a first data point that is based in part on a combination of at least two data points of the input irregular time series; and

use the plurality of variant time series for a machine learning task.

14. The system of claim 13 , wherein the machine learning task trains a neural network model and the machine learning service includes further instructions that upon execution cause the machine learning service to:

for each variant time series of the plurality of variant time series, process the variant time series with the neural network model to generate an output; and

update a parameter of the neural network model based at least in part on the output.

15. The system of claim 13 , wherein the machine learning task generates a prediction vector, and the machine learning service includes further instructions that upon execution cause the machine learning service to:

generate a first output from a neural network model based on a first variant time series of the plurality of variant time series;

generate a second output from the neural network model based on a second variant time series of the plurality of variant time series; and

output the prediction vector based in part on the first output and the second output.

16. The system of claim 15 , wherein the first output and the second output are scaled by a weight.

17. The system of claim 13 , wherein to generate the first variant time series the machine learning service includes further instructions that upon execution cause the machine learning service to:

group the data points in the input irregular time series into a plurality of clusters, wherein a number of clusters in the plurality of clusters is less than the number of data points in the input irregular time series, wherein each cluster corresponds to a data point in the first variant time series of the plurality of variant time series;

generate, for each cluster, a time stamp from the time stamps of the data points grouped into the cluster; and

generate, for each cluster, a feature vector from the feature vectors of the data points grouped into the cluster.

18. The system of claim 13 , wherein to generate the first variant time series the machine learning service includes further instructions that upon execution cause the machine learning service to:

select a plurality of edges in the input irregular time series, an edge corresponding to a time between consecutive data points in the input irregular time series;

generate, for each selected edge, a time stamp from the time stamps of the consecutive data points bounding the selected edge; and

generate, for each selected edge, a feature vector from the feature vectors of the consecutive data points bounding the selected edge.

19. The system of claim 13 , wherein to generate the first variant time series the machine learning service includes further instructions that upon execution cause the machine learning service to:

generate a plurality of clusters, each cluster corresponding to a regularly spaced timestamp between an earliest timestamp of a second data point in the input irregular time series and a latest timestamp of a third data point in the input irregular time series, wherein a number of clusters in the plurality of clusters is less than the number of data points in the input irregular time series;

associate each data point in the input irregular time series with one of the plurality of clusters;

generate, for each cluster, a time stamp from the time stamps of the data points associated with the cluster; and

generate, for each cluster, a feature vector from the feature vectors of the data points associated with the cluster.

20. The system of claim 13 , wherein the first data point of the first variant time series includes an indication of a number of data points of the input irregular time series that were combined to form the first data point.