Ordinal time series classification with missing information
A method classifies missing labels. The method computes, using a neural network model trained on training data, rank-based statistics of a feature of a time series segment to attempt to select two candidate labels from the training data that the segment most likely belongs to. The method classifies the segment using k-NN-based classification applied to the training data, responsive to the two candidate labels being present in the training data. The method classifies the segment by hypothesis testing, responsive to only one candidate label being present in the training data. The method classifies the segment into a class with higher values of the rank-based statistics from among a plurality of classes with different values of the rank-based statistics, responsive to no candidate labels being present in the training data. The method corrects a prediction by an applicable one of the classifying steps by majority voting with time windows.
1. A computer-implemented method for time series classification of missing labels, comprising:
extracting a feature of an incoming time series segment to be classified during an inference stage;
computing, by a hardware processor using a neural network model trained on training data, rank-based statistics of the feature to attempt to select two candidate labels from the training data that the incoming time series segment most likely belongs to;
classifying the incoming time series segment using k-NN-based classification applied to the training data, responsive to the two candidate labels being present in the training data;
classifying the incoming time series segment by hypothesis testing, responsive to only one of the two candidate labels being present in the training data;
classifying the incoming time series segment into a class with higher values of the rank-based statistics from among a plurality of classes with different values of the rank-based statistics, responsive to none of the two candidate labels being present in the training data; and
correcting a prediction by an applicable one of the classifying steps by majority voting with time windows.
2. The computer-implemented method of claim 1 , further comprising training a neural network model on the training data based on ordinal quadruplet loss.
3. The computer-implemented method of claim 2 , wherein the ordinal quadruplet loss comprises a similarity component, a discrimination component, and a feature order component with respect to a feature space.
4. The computer-implemented method of claim 2 , wherein the ordinal quadruplet loss comprises a triplet loss and a log ratio loss.
5. The computer-implemented method of claim 1 , further comprising computing, for each of the two candidate labels a label retrieval vector from the each of the two candidate labels to each of present labels of the original data.
6. The computer-implemented method of claim 1 , further comprising computing, for each incoming times series to be classified during the inference stage, a test retrieval vector from each of the two candidate labels to a feature center of each of present labels in a feature space.
7. The computer-implemented method of claim 6 , wherein the feature center is computed as an average of features of incoming time series segments having a same label.
8. The computer-implemented method of claim 1 , wherein the hypothesis testing comprises determining outliers for a missing class using statistical methods.
9. The computer-implemented method of claim 8 , wherein determining outliers comprises determining a distribution of distances from the training data in present classes to a class center of an evaluated class with respect to a threshold distance.
10. The computer-implemented method of claim 1 , wherein said detecting step comprises making label predictions within a window of time using a majority voting scheme that selects a most predicted class during the window of time.
11. The computer-implemented method of claim 1 , further comprising replacing an impending failing workplace machine with a backup workplace machine responsive to the prediction to avoid system downtime.
12. A computer program product for time series classification of missing labels, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method comprising:
extracting, by a hardware processor, a feature of an incoming time series segment to be classified during an inference stage;
computing, by the hardware processor using a neural network model trained on training data, rank-based statistics of the feature to attempt to select two candidate labels from the training data that the incoming time series segment most likely belongs to;
classifying, by the hardware processor, the incoming time series segment using k-NN-based classification applied to the training data, responsive to the two candidate labels being present in the training data;
classifying the incoming time series segment by hypothesis testing, responsive to only one of the two candidate labels being present in the training data;
classifying the incoming time series segment into a class with higher values of the rank-based statistics from among a plurality of classes with different values of the rank-based statistics, responsive to none of the two candidate labels being present in the training data; and
correcting a prediction by an applicable one of the classifying steps by majority voting with time windows.
13. The computer program product of claim 12 , wherein the method further comprises training a neural network model on the training data based on ordinal quadruplet loss.
14. The computer program product of claim 13 , wherein the ordinal quadruplet loss comprises a similarity component, a discrimination component, and a feature order component with respect to a feature space.
15. The computer program product of claim 13 , wherein the ordinal quadruplet loss comprises a triplet loss and a log ratio loss.
16. The computer program product of claim 12 , wherein the method further comprises computing, for each of the two candidate labels a label retrieval vector from the each of the two candidate labels to each of present labels of the original data.
17. The computer program product of claim 12 , wherein the method further comprises computing, for each incoming times series to be classified during the inference stage, a test retrieval vector from each of the two candidate labels to a feature center of each of present labels in a feature space.
18. The computer program product of claim 17 , wherein the feature center is computed as an average of features of incoming time series segments having a same label.
19. The computer program product of claim 12 , wherein the hypothesis testing comprises determining outliers for a missing class using statistical methods.
20. A computer processing system for time series classification of missing labels, comprising:
a memory device for storing program code; and
a hardware processor operatively coupled to the memory device for storing the program code to
extract a feature of an incoming time series segment to be classified during an inference stage;
compute, using a neural network model trained on training data, rank-based statistics of the feature to attempt to select two candidate labels from the training data that the incoming time series segment most likely belongs to;
classify the incoming time series segment using k-NN-based classification applied to the training data, responsive to the two candidate labels being present in the training data;
classify the incoming time series segment by hypothesis testing, responsive to only one of the two candidate labels being present in the training data;
classify the incoming time series segment into a class with higher values of the rank-based statistics from among a plurality of classes with different values of the rank-based statistics, responsive to none of the two candidate labels being present in the training data; and
correct a prediction by an applicable one of the classifying steps by majority voting with time windows.