Validation of autoencoders that perform timestamp compression for printing system data
Systems and methods are provided for validating an autoencoder for a printing system. In one embodiment, a system stores values associated with individual features at a printing system and further stores times indicating when the values were determined at the printing system. The system also operates an autoencoder to perform time-compression upon the values and the times. Validating the autoencoder includes: determining predictive powers of the individual features prior to time-compression via partial F tests, determining predictive powers of the individual features after time-compression via partial F tests, and determining differences between the predictive powers of the individual features before time-compression and the predictive powers of the individual features after time-compression. Depending upon differences in the predictive powers, a report is generated indicating validity or invalidity of the autoencoder.
1 . A system for validating an autoencoder for a printing system, the system comprising:
a memory storing values associated with individual features at a printing system and further storing times indicating when the values were determined at the printing system; and
a controller configured to operate an autoencoder to perform time-compression upon the values and the times,
the controller further configured to validate the autoencoder by: determining predictive powers of the individual features prior to time-compression via partial F tests, determining predictive powers of the individual features after time-compression via partial F tests, and determining differences between the predictive powers of the individual features before time-compression and the predictive powers of the individual features after time-compression,
the controller further configured to generate a report for a user indicating invalidity of the autoencoder, upon determining that the differences in the predictive powers are not within a threshold, and
the controller further configured to generate a report for the user indicating validity of the autoencoder, upon determining that the differences in the predictive powers are within the threshold.
2 . The system of claim 1 wherein:
the controller is further configured to use the autoencoder to train a machine learning model, upon determining that the differences in the predictive powers are within the threshold.
3 . The system of claim 1 wherein:
each partial F test produces multiple sets of f-values and p-values, and the sets are compared to determine the differences in the predictive powers.
4 . The system of claim 1 wherein:
the controller is further configured to perform a partial F test by determining prediction accuracy for a machine learning model after training with a group of individual features, determining prediction accuracy for the machine learning model after training with a reduced group wherein a selected feature has been removed, and identifying a difference between the prediction accuracies as a predictive power of the selected feature that has been removed.
5 . The system of claim 4 wherein:
the individual features comprise between ten and one hundred different features, and each group comprises between three and five different features.
6 . The system of claim 1 , wherein:
the controller is further configured to determine predictive power of the individual features by iteratively performing partial F tests upon different, randomly selected, groups of the individual features.
7 . The system of claim 1 , wherein:
the autoencoder performs time-compression by estimating values associated with the features at a set of predetermined times, based on the values stored in the memory.
8 . The system of claim 1 wherein:
the features are selected from the group consisting of: fan speed for a printer, print medium temperature, fan temperature, print medium speed, print medium alignment, and ink consumption.
9 . A non-transitory computer readable medium embodying programmed instructions which, when executed by a processor, are operable for:
storing values associated with individual features at a printing system and times indicating when the values were determined at the printing system;
operating an autoencoder to perform time-compression upon the values and the times;
validating the autoencoder by:
determining predictive powers of the individual features prior to time-compression via partial F tests;
determining predictive powers of the individual features after time-compression via partial F tests; and
determining differences between the predictive powers of the individual features before time-compression and the predictive powers of the individual features after time-compression;
generating a report for a user indicating invalidity of the autoencoder, upon determining that the differences in the predictive powers are not within a threshold; and
generating a report for the user indicating validity of the autoencoder, upon determining that the differences in the predictive powers are within the threshold.
10 . The medium of claim 9 , wherein the instructions are further operable for:
using the autoencoder to train a machine learning model, upon determining that the differences in the predictive powers are within the threshold.
11 . The medium of claim 9 wherein:
each partial F test produces multiple sets of f-values and p-values, and the sets are compared to determine the differences in the predictive powers.
12 . The medium of claim 9 , wherein:
performing a partial F test comprises:
determining prediction accuracy for a machine learning model after training with a group of individual features;
determining prediction accuracy for the machine learning model after training with a reduced group wherein a selected feature has been removed; and
identifying a difference between the prediction accuracies as a predictive power of the selected feature that has been removed.
13 . The medium of claim 12 , wherein:
the individual features comprise between ten and one hundred different features, and each group comprises between three and five different features.
14 . The medium of claim 9 , wherein:
determining predictive power of the individual features comprises iteratively performing partial F tests upon different, randomly selected, groups of the individual features.
15 . The medium of claim 9 , wherein:
operating the autoencoder to perform time-compression comprises estimating values associated with the features at a set of predetermined times, based on the values stored in memory.
16 . A method for validating an autoencoder for a printing system, the method comprising:
storing values associated with individual features at the printing system and times indicating when the values were determined at the printing system;
operating the autoencoder to perform time-compression upon the values and the times;
validating the autoencoder by:
determining predictive powers of the individual features prior to time-compression via partial F tests;
determining predictive powers of the individual features after time-compression via partial F tests; and
determining differences between the predictive powers of the individual features before time-compression and the predictive powers of the individual features after time-compression;
generating a report for a user indicating invalidity of the autoencoder, upon determining that the differences in the predictive powers are not within a threshold; and
generating a report for the user indicating validity of the autoencoder, upon determining that the differences in the predictive powers are within the threshold.
17 . The method of claim 16 further comprising:
using the autoencoder to train a machine learning model, upon determining that the differences in the predictive powers are within the threshold.
18 . The method of claim 16 wherein:
each partial F test produces multiple sets of f-values and p-values, and the sets are compared to determine the differences in the predictive powers.
19 . The method of claim 16 wherein:
performing a partial F test comprises:
determining prediction accuracy for a machine learning model after training with a group of individual features;
determining prediction accuracy for the machine learning model after training with a reduced group wherein a selected feature has been removed; and
identifying a difference between the prediction accuracies as a predictive power of the selected feature that has been removed.
20 . The method of claim 19 wherein:
the individual features comprise between ten and one hundred different features, and each group comprises between three and five different features.