Explainers for machine learning classifiers

Abstract

A transformed data set corresponding to a machine learning classifier's training data set is generated. Each transformed record contains a modified version of a corresponding training record, as well as the prediction made for the training record by the classifier. A set of explanatory rules is minded from the transformed data set, with each rule indicating a relationship between the prediction and one or more features corresponding to the training records. From among the rule set, a particular matching rule is selected to provide an easy-to-understand explanation for a prediction made by the classifier for an observation record which is not part of the training set.

1 .- 20 . (canceled) 
     
     
         21 . A computer-implemented method, comprising:
 performing, at a cloud based machine learning service:
 obtaining an indication of at least a portion of a training data set of a machine learning model; 
 identifying, using at least the portion of the training data set, one or more rules of an explainer for at least some predictions produced by the machine learning model; and 
 providing, using the explainer, an explanation for a first prediction produced by the machine learning model with respect to at least a first input record. 
   
     
     
         22 . The computer-implemented method as recited in  claim 21 , wherein the one or more rules comprise a plurality of rules, the computer-implemented method further comprising performing, at the cloud based machine learning service:
 ranking the plurality of rules based on one or more criteria; and   selecting, for the explanation, the first rule from the plurality of rules based at least in part on a result of the ranking.   
     
     
         23 . The computer-implemented method as recited in  claim 22 , wherein a criterion of the one or more criteria is based at least in part on one or more of: (a) a support metric indicative, with respect to the first rule, of a number of observation records whose attribute values match a predicate defined in the first rule, or (b) a confidence metric indicative, with respect to the first rule, of an accuracy of the first explanatory rule. 
     
     
         24 . The computer-implemented method as recited in  claim 21 , wherein a first rule of the one or more rules is expressed in terms of a first transformation of a first attribute of input records of the machine learning model, the computer-implemented method further comprising performing, at the cloud based machine learning service:
 applying a reverse transformation, with respect to the first transformation, on an internal representation of the first attribute of the first input record, wherein the explanation comprises a result of the reverse transformation.   
     
     
         25 . The computer-implemented method as recited in  claim 21 , wherein the one or more rules comprise a plurality of rules, the computer-implemented method further comprising performing, at the cloud based machine learning service:
 selecting, for the explanation, the first rule from the plurality of rules based at least in part on a number of predicates included in the first rule.   
     
     
         26 . The computer-implemented method as recited in  claim 21 , further comprising performing, at the cloud based machine learning service:
 receiving a set of predictions generated by the machine learning model for the portion of the training data set, wherein identifying the one or more rules comprises utilizing the set of predictions.   
     
     
         27 . The computer-implemented method as recited in  claim 21 , wherein the machine learning model utilizes one or more of: (a) a neural network algorithm, (b) a random forest algorithm, or (c) a boosted gradient tree algorithm. 
     
     
         28 . A system, comprising:
 one or more computing devices;   wherein the one or more computing devices include instructions that upon execution on or across the one or more computing devices cause the one or more computing devices to:
 obtain an indication of at least a portion of a training data set of a machine learning model; 
 identify, using at least the portion of the training data set, one or more rules of an explainer for at least some predictions produced by the machine learning model; and 
 provide, using the explainer, an explanation for a first prediction produced by the machine learning model with respect to at least a first input record. 
   
     
     
         29 . The system as recited in  claim 28 , wherein the one or more rules comprise a plurality of rules, and wherein the one or more computing devices include further instructions that upon execution on or across the one or more computing devices further cause the one or more computing devices to:
 rank the plurality of rules based on one or more criteria; and   select, for the explanation, a first rule from the plurality of rules based at least in part on a result of the ranking.   
     
     
         30 . The system as recited in  claim 29 , wherein a criterion of the one or more criteria is based at least in part on one or more of: (a) a support metric indicative, with respect to the first rule, of a number of observation records whose attribute values match a predicate defined in the first rule, or (b) a confidence metric indicative, with respect to the first rule, of an accuracy of the first explanatory rule. 
     
     
         31 . The system as recited in  claim 28 , wherein a first rule of the one or more rules is expressed using a first transformation of a first attribute of input records of the machine learning model, and wherein the one or more computing devices include further instructions that upon execution on or across the one or more computing devices further cause the one or more computing devices to:
 apply a reverse transformation, with respect to the first transformation, on an internal representation of the first attribute of the first input record, wherein the explanation comprises a result of the reverse transformation.   
     
     
         32 . The system as recited in  claim 28 , wherein the one or more rules comprise a plurality of rules, and wherein the one or more computing devices include further instructions that upon execution on or across the one or more computing devices further cause the one or more computing devices to:
 select, for the explanation, the first rule from the plurality of rules based at least in part on a number of attribute predicates included in the first rule.   
     
     
         33 . The system as recited in  claim 28 , wherein the one or more computing devices include further instructions that upon execution on or across the one or more computing devices further cause the one or more computing devices to:
 receive a set of predictions generated by the machine learning model for the portion of the training data set, wherein the one or more rules are identified based at least in part on the set of predictions.   
     
     
         34 . The system as recited in  claim 28 , wherein the machine learning model comprises a classification model. 
     
     
         35 . One or more non-transitory computer-accessible storage media storing program instructions that when executed on or across one or more processors cause the one or more processors to:
 obtain an indication of at least a portion of a training data set of a machine learning model;   generate, using at least the portion of the training data set, at least a portion of an explainer of at least some predictions produced by the machine learning model; and   provide, using the explainer, an explanation for a first prediction produced by the machine learning model with respect to at least a first input record.   
     
     
         36 . The one or more non-transitory computer-accessible storage media as recited in  claim 35 , wherein the explainer comprises a plurality of rules, the one or more non-transitory computer-accessible storage media storing further program instructions that when executed on or across one or more processors cause the one or more processors to:
 rank the plurality of rules based on one or more criteria; and   select, for the explanation, a first rule from the plurality of rules based at least in part on a result of the ranking.   
     
     
         37 . The one or more non-transitory computer-accessible storage media as recited in  claim 36 , wherein a criterion of the one or more criteria is based at least in part on one or more of: (a) a support metric indicative, with respect to the first rule, of a number of observation records whose attribute values match a predicate defined in the first rule, or (b) a confidence metric indicative, with respect to the first rule, of an accuracy of the first explanatory rule. 
     
     
         38 . The one or more non-transitory computer-accessible storage media as recited in  claim 35 , wherein a rule of the explainer is expressed in terms of a first transformation of a first attribute of input records of the machine learning model, the one or more non-transitory computer-accessible storage media storing further program instructions that when executed on or across one or more processors cause the one or more processors to:
 apply a reverse transformation, with respect to the first transformation, on an internal representation of the first attribute of the first input record, wherein the explanation comprises a result of the reverse transformation.   
     
     
         39 . The one or more non-transitory computer-accessible storage media as recited in  claim 35 , wherein the explainer comprises a plurality of rules, the one or more non-transitory computer-accessible storage media storing further program instructions that when executed on or across one or more processors cause the one or more processors to:
 select, for the explanation, the first rule from the plurality of rules based at least in part on a number of attribute predicates included in the first rule.   
     
     
         40 . The one or more non-transitory computer-accessible storage media as recited in  claim 35 , storing further program instructions that when executed on or across one or more processors cause the one or more processors to:
 obtain a set of predictions generated by the machine learning model for the portion of the training data set, wherein the portion of the explainer is generated based at least in part on the set of predictions.