Dynamic user-interface comparison between machine learning output and training data
Systems/techniques that facilitate dynamic user-interface comparison between machine learning output and training data are provided. In various embodiments, a system can access a data candidate on which a machine learning classifier is to be executed. In various aspects, the system can generate, via execution of the machine learning classifier, a predicted classification based on the data candidate. In various instances, the system can render, on an electronic display, a confidence graph that visually depicts how well or how poorly the data candidate and the predicted classification fit into an annotated training dataset on which the machine learning classifier has been trained.
1 . A system, comprising:
a memory configured to store computer-executable components; and
a processor configured to execute at least one of the computer-executable components that
accesses a data candidate on which a machine learning classifier is to be executed;
generates, via execution of the machine learning classifier, a predicted classification based on the data candidate, wherein the machine learning classifier receives as input the data candidate, and wherein the machine learning classifier produces as output the predicted classification; and
renders, on an electronic display, a confidence graph that visually depicts how well or how poorly the data candidate and the predicted classification fit into an annotated training dataset on which the machine learning classifier has been trained, wherein the confidence graph comprises at least one feature-based coordinate axis that respectively corresponds to at least one data feature, wherein at least some portion of the annotated training dataset is plotted in the confidence graph along the at least one feature-based coordinate axis and is color-coded or symbol-coded according to respective ground-truth classifications of a group of ground-truth classifications, wherein the data candidate is plotted in the confidence graph along the at least one feature-based coordinate axis and is color-coded or symbol-coded according to a ground-truth classification of the group of ground-truth classifications corresponding to the predicted classification, and wherein the confidence graph concurrently depicts a visual representation of:
the ground-truth classification corresponding to the predicted classification, and
for respective data feature of the at least one data feature, a set of overlapping respective feature distribution representations of feature distributions for the respective ground-truth classifications of the group of ground-truth classifications.
2 . The system of claim 1 , wherein the confidence graph concurrently depicts the visual representation further of relative distances between the data candidate and respective subsets of training data candidates of the at least some portion of the annotated training dataset associated with each ground-truth classification of the group of ground-truth classifications.
3 . The system of claim 2 , wherein the relative distances between the data candidate and the respective subsets of training data candidates are depicted via a scatterplot, and wherein the respective feature distribution representations are color-coded or symbol-coded according to the respective ground-truth classifications.
4 . The system of claim 3 , wherein the confidence graph indicates, via the set of overlapping respective feature distribution representations of the feature distributions for the respective data feature and the respective ground-truth classifications, which of the feature distributions for the respective data feature has a mean that is nearest to the data candidate.
5 . The system of claim 1 , wherein the confidence graph comprises at least one of a one-dimensional histogram, a two-dimensional scatterplot, a three-dimensional scatterplot, a heat map, a density map, or a t-distributed stochastic neighbor embedding plot.
6 . The system of claim 1 , wherein the at least one of the computer-executable components further:
selects the at least some portion of the annotated training dataset based on user-provided input.
7 . The system of claim 1 , wherein the at least one of the computer-executable components further:
selects the at least some portion of the annotated training dataset based on the data candidate.
8 . The system of claim 1 , wherein the data candidate is a medical image that depicts an anatomical structure of a medical patient, wherein the predicted classification is a pathology diagnosis, and wherein the at least one data feature comprises at least one of a size metric or a shape metric associated with the anatomical structure.
9 . A computer-implemented method, comprising: accessing, by a system operatively coupled to a processor, a data candidate on which a machine learning classifier is to be executed;
generating, by the system and via execution of the machine learning classifier, a predicted classification based on the data candidate, wherein the machine learning classifier receives as input the data candidate, and wherein the machine learning classifier produces as output the predicted classification; and
rendering, by the system and on an electronic display, a confidence graph that visually depicts how well or how poorly the data candidate and the predicted classification fit into an annotated training dataset on which the machine learning classifier has been trained, wherein the confidence graph comprises at least one feature-based coordinate axis that respectively corresponds to at least one data feature, wherein at least some portion of the annotated training dataset is plotted in the confidence graph along the at least one feature-based coordinate axis and is color-coded or symbol-coded according to respective ground-truth classifications of a group of ground-truth classifications, wherein the data candidate is plotted in the confidence graph along the at least one feature-based coordinate axis and is color-coded or symbol-coded according to a ground-truth classification of the group of ground-truth classifications corresponding to the predicted classification, and wherein the confidence graph concurrently depicts a visual representation of:
the ground-truth classification corresponding to the predicted classification, and
for respective data feature of the at least one data feature, a set of overlapping respective feature distribution representations of feature distributions for the respective ground-truth classifications of the group of ground-truth classifications.
10 . The computer-implemented method of claim 9 , wherein the confidence graph concurrently depicts the visual representation further of relative distances between the data candidate and respective subsets of training data candidates of the at least some portion of the annotated training dataset associated with each ground-truth classification of the group of ground-truth classifications.
11 . The computer-implemented method of claim 10 , wherein the relative distances between the data candidate and the respective subsets of training data candidates are depicted via a scatterplot, and wherein the feature distribution representations are color-coded or symbol-coded according to the respective ground-truth classifications.
12 . The computer-implemented method of claim 11 , wherein the confidence graph indicates, via the set of overlapping respective feature distribution representations of the feature distributions for the respective data feature and the respective ground-truth classifications, which of the feature distributions for the respective data feature has a mean that is nearest to the data candidate.
13 . The computer-implemented method of claim 9 , wherein the confidence graph comprises at least one of a one-dimensional histogram, a two-dimensional scatterplot, a three-dimensional scatterplot, a heat map, a density map, or a t-distributed stochastic neighbor embedding plot.
14 . The computer-implemented method of claim 9 , further comprising:
selecting, by the system, the at least some portion of the annotated training dataset based on user-provided input.
15 . The computer-implemented method of claim 9 , further comprising:
selecting, by the system, the at least some portion of the annotated training dataset based on the data candidate.
16 . The computer-implemented method of claim 9 , wherein the data candidate is a medical image that depicts an anatomical structure of a medical patient, wherein the predicted classification is a pathology diagnosis, and wherein the at least one data feature comprises at least one of a size metric or a shape metric associated with the anatomical structure.
17 . A computer program product for facilitating dynamic user-interface comparison between machine learning output and training data, the computer program product comprising a computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
access a data candidate on which a machine learning classifier is to be executed;
generate, via execution of the machine learning classifier, a predicted classification based on the data candidate, wherein the machine learning classifier receives as input the data candidate, and wherein the machine learning classifier produces as output the predicted classification; and
render, on an electronic display, a confidence graph that visually depicts how well or how poorly the data candidate and the predicted classification fit into an annotated training dataset on which the machine learning classifier has been trained, wherein the confidence graph comprises at least one feature-based coordinate axis that respectively corresponds to at least one data feature, wherein at least some portion of the annotated training dataset is plotted in the confidence graph along the at least one feature-based coordinate axis and is color-coded or symbol-coded according to respective ground-truth classifications of a group of ground-truth classifications, wherein the data candidate is plotted in the confidence graph along the at least one feature-based coordinate axis and is color-coded or symbol-coded according to a ground-truth classification of the group of ground-truth classifications corresponding to the predicted classification, and wherein the confidence graph concurrently depicts a visual representation of:
the ground-truth classification corresponding to the predicted classification, and
for respective data feature of the at least one data feature, a set of overlapping respective feature distribution representations of feature distributions for the respective ground-truth classifications of the group of ground-truth classifications.
18 . The computer program product of claim 17 , wherein the confidence graph concurrently depicts the visual representation further of relative distances between the data candidate and respective subsets of training data candidates of the at least some portion of the annotated training dataset associated with each ground-truth classification of the group of ground-truth classifications.
19 . The computer program product of claim 18 , wherein the relative distances between the data candidate and the respective subsets of training data candidates are depicted via a scatterplot, and wherein the respective feature distribution representations are color-coded or symbol-coded according to the respective ground-truth classifications.
20 . The computer program product of claim 17 , wherein the confidence graph comprises at least one of a one-dimensional histogram, a two-dimensional scatterplot, a three-dimensional scatterplot, a heat map, a density map, or a t-distributed stochastic neighbor embedding plot.