CURRENT USAGE ESTIMATION FOR ELECTRONIC DEVICES

1 . A method of estimating current consumption of an electronic device, the method comprising:

obtaining log data comprising a record of at least some activities of the electronic device during a selected time period;

parsing the log data into a plurality of component digests;

estimating current consumption values for the component digests based on component signatures of the component digests, a current consumption model and a machine learning technique; and

processing the estimated current consumption values to estimate current consumption data for the electronic device.

2 . The method of claim 1 , wherein estimating the current consumption value for a given component digest comprises:

determining the component signature of the given component digest;

locating a training signature from training set data that corresponds to the component signature; and

setting the estimated current consumption value to a current label associated with the located training signature.

3 . The method of claim 2 , wherein determining the component signature of the given component digest comprises mapping the component digest to an N-dimensional vector using a transformation that is at least approximate distance preserving.

4 . The method of claim 3 , wherein the mapping comprises mapping the component digest to a bag-of-word array by counting occurrences of text in the component digest and applying different weights to occurrences of different text in the component digest and then mapping the bag-of-word array to the component signature using the at least approximate distance preserving transformation.

5 . The method of claim 1 , wherein the method comprises estimating the current consumption model by:

obtaining training log data when executing a plurality of activities on the electronic device, the training log data comprising kernel event log data and current log data;

synchronizing the current log data with the kernel event log data when the current log data and the kernel event log data are not synchronized;

merging the kernel event log data with the current log data to form kernel-event-current log data;

parsing the kernel-event-current-log data into a plurality of raw digests;

determining training signatures for the raw digests after removing the current log values from the raw digests;

determining labels for the training signatures from the current usage values of the raw digests that correspond to the training signatures; and

applying the machine learning algorithm to the training signatures and corresponding labels to determine a relationship between the activities and the current usage used to generate the current consumption model.

6 . The method of claim 5 , wherein determining a label associated with a given raw digest further comprises:

extracting the current consumption values from the given raw digest;

computing an average current consumption value from averaging the current consumption values over the length of the given raw digest; and

assigning the average current consumption value as the label associated with the training signature corresponding to the given raw digest.

7 . The method of claim 1 , wherein the machine learning algorithm is one of a VP-tree based nearest-neighbor algorithm, a support vector machine algorithm, a decision-tree algorithm, a linear or logistic regression algorithm, a boosted decision tree and a neural network.

8 . An electronic device comprising:

a plurality of subsystems for providing various functions;

an operating system for enabling execution of software applications and operation of the subsystems as well as logging activity of the electronic device in log data;

a processor that controls the operation of the device, the processor being configured to estimate current consumption of the electronic device by obtaining log data comprising a record of at least some activities of the electronic device during a selected time period; parsing the log data into a plurality of component digests; estimating current consumption values for the component digests based on component signatures of the component digests, a current consumption model and a machine learning technique; and

processing the estimated current consumption values to estimate current consumption data for the electronic device.

9 . The device of claim 8 , wherein estimating the current consumption value for a given component digest comprises:

determining the component signature of the given component digest;

locating a training signature from training set data that corresponds to the component signature; and

setting the estimated current consumption value to a current label associated with the located training signature.

10 . The device of claim 9 , wherein determining the component signature of the given component digest comprises mapping the component digest to an N-dimensional vector using a transformation that is at least approximate distance preserving.

11 . The device of claim 10 , wherein the mapping comprises mapping the component digest to a bag-of-word array by counting occurrences of text in the component digest and applying different weights to occurrences of different text in the component digest and then mapping the bag-of-word array to the component signature using the at least approximate distance preserving transformation.

12 . The device of claim 8 , wherein the method comprises estimating the current consumption model by:

obtaining training log data when executing a plurality of activities on the electronic device, the training log data comprising kernel event log data and current log data;

synchronizing the current log data with the kernel event log data when the current log data and the kernel event log data are not synchronized;

merging the kernel event log data with the current log data to form kernel-event-current log data;

parsing the kernel-event-current-log data into a plurality of raw digests;

determining training signatures for the raw digests after removing the current log values from the raw digests;

determining labels for the training signatures from the current usage values of the raw digests that correspond to the training signatures; and

applying the machine learning algorithm to the training signatures and corresponding labels to determine a relationship between the activities and the current usage used to generate the current consumption model.

13 . The device of claim 12 , wherein determining a label associated with a given raw digest further comprises:

extracting the current consumption values from the given raw digest;

computing an average current consumption value from averaging the current consumption values over the length of the given raw digest; and

assigning the average current consumption value as the label associated with the training signature corresponding to the given raw digest.

14 . The device of claim 8 , wherein the machine learning algorithm is one of a VP-tree based nearest-neighbor algorithm, a support vector machine algorithm, a decision-tree algorithm, a linear or logistic regression algorithm, a boosted decision tree and a neural network.

15 . A computer readable medium comprising a plurality of instructions executable on a microprocessor of an electronic device for adapting the electronic device to implement a method of estimating current consumption for the electronic device, wherein the method comprises:

obtaining log data comprising a record of at least some activities of the electronic device during a selected time period;

parsing the log data into a plurality of component digests;

estimating current consumption values for the component digests based on component signatures of the component digests, a current consumption model and a machine learning technique; and

processing the estimated current consumption values to estimate current consumption data for the electronic device.

16 . The computer readable medium of claim 15 , wherein estimating the current consumption value for a given component digest comprises:

determining the component signature of the given component digest;

locating a training signature from training set data that corresponds to the component signature; and

setting the estimated current consumption value to a current label associated with the located training signature.

17 . The computer readable medium of claim 16 , wherein determining the component signature of the given component digest comprises mapping the component digest to an N-dimensional vector using a transformation that is at least approximate distance preserving.

18 . The computer readable medium of claim 17 , wherein the mapping comprises mapping the component digest to a bag-of-word array by counting occurrences of text in the component digest and applying different weights to occurrences of different text in the component digest and then mapping the bag-of-word array to the component signature using the at least approximate distance preserving transformation.

19 . The computer readable medium of claim 15 , wherein the method comprises estimating the current consumption model by:

obtaining training log data when executing a plurality of activities on the electronic device, the training log data comprising kernel event log data and current log data;

synchronizing the current log data with the kernel event log data when the current log data and the kernel log data are not synchronized;

merging the kernel event log data with the current log data to form kernel-event-current log data;

parsing the kernel-event-current-log data into a plurality of raw digests;

determining training signatures for the raw digests after removing the current log values from the raw digests;

determining labels for the training signatures from the current usage values of the raw digests that correspond to the training signatures; and

applying the machine learning algorithm to the training signatures and corresponding labels to determine a relationship between the activities and the current usage used to generate the current consumption model.

20 . The computer readable medium of claim 19 , wherein determining a label associated with a given raw digest further comprises:

extracting the current consumption values from the given raw digest;

computing an average current consumption value from averaging the current consumption values over the length of the given raw digest; and

assigning the average current consumption value to be the label associated with the training signature corresponding to the given raw digest.

21 . The computer readable medium of claim 15 , wherein the machine learning algorithm is one of a VP-tree based nearest-neighbor algorithm, a support vector machine algorithm, a decision-tree algorithm, a linear or logistic regression algorithm, a boosted decision tree and a neural network.