Machine learning techniques for validating and mutating outputs from predictive systems
Various embodiments of the present disclosure provide methods, apparatus, systems, computing devices, computing entities, and/or the like for revising classifier predictions, wherein classification labels are tokenized according to an index of predefined classification labels. An embedding is created for each token using an embedding model. The embeddings are provided to a machine learning model that accepts a sequence of tokens as input and produces a sequence of tokens as output. Tokens are extracted from the output of the machine learning model and a classification label corresponding to each token is retrieved according to the index of predefined classification labels.
1 . A computer-implemented method comprising:
receiving, by a computing device and originating from a predictive system, predictive output data, the predictive output data comprising a plurality of classification labels;
generating, by the computing device, an input sequence of tokens, the input sequence of tokens comprising a tokenization of the plurality of classification labels based at least in part on a numerical index of predefined classification labels;
generating, by the computing device and using a prediction validation machine learning framework, a mutation of the input sequence of tokens, wherein:
the prediction validation machine learning framework comprises an embedding machine learning model and a mutation machine learning model, wherein:
(i) the embedding machine learning model is configured to (a) generate, for each token of the input sequence of tokens, a token embedding, and (b) concatenate each of the token embedding into a token embedding matrix,
(ii) the mutation machine learning model is configured to generate an output matrix, comprising an output sequence of tokens, based at least in part on the token embedding matrix,
(iii) the mutation of the input sequence of tokens comprises retrieval of one or more output classification labels associated with the output sequence of tokens based at least in part on the numerical index of predefined classification labels; and
performing, by the computing device, one or more prediction-based actions based at least in part on the mutation of the input sequence of tokens.
2 . The computer-implemented method of claim 1 , wherein training the mutation machine learning model comprises imparting the mutation machine learning model with a plurality of training sequence of tokens pairs, each training sequence of tokens pair comprising a training input sequence of tokens and a training output sequence of tokens.
3 . The computer-implemented method of claim 1 , wherein the mutation machine learning model comprises a sequence-to-sequence machine learning model.
4 . The computer-implemented method of claim 1 , wherein the token embedding matrix comprises a first dimension associated with an embedding size and a second dimension associated with a quantity of tokens associated with the input sequence of tokens.
5 . The computer-implemented method of claim 1 , wherein the output matrix comprises a first dimension associated with a quantity of possible predictions based at least in part on the numerical index of predefined classification labels and a second dimension associated with a quantity of predictions associated with the output sequence of tokens.
6 . The computer-implemented method of claim 5 , wherein the output sequence of tokens is based at least in part on position values in the first dimension including a highest value for each position in the second dimension.
7 . The computer-implemented method of claim 6 , the one or more output classification labels are based at least in part on integer values in the numerical index of predefined classification labels.
8 . The computer-implemented method of claim 1 , wherein the numerical index of predefined classification labels comprises a set of classification labels mapped to integer values.
9 . A system comprising
one or more processors and
one or more non-transitory computer readable media storing processor-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
receiving predictive output data from a predictive system, the predictive output data comprising a plurality of classification labels;
generating an input sequence of tokens, the input sequence of tokens comprising a tokenization of the plurality of classification labels based at least in part on a numerical index of predefined classification labels;
generating, using a prediction validation machine learning framework, a mutation of the input sequence of tokens, wherein:
the prediction validation machine learning framework comprises an embedding machine learning model and a mutation machine learning model, wherein:
(i) the embedding machine learning model is configured to (a) generate, for each token of the input sequence of tokens, a token embedding, and (b) concatenate each of the token embedding into a token embedding matrix,
(ii) the mutation machine learning model is configured to generate an output matrix, comprising an output sequence of tokens, based at least in part on the token embedding matrix,
(iii) the mutation of the input sequence of tokens comprises retrieval of one or more output classification labels associated with the output sequence of tokens based at least in part on the numerical index of predefined classification labels; and
performing one or more prediction-based actions based at least in part on the mutation of the input sequence of tokens.
10 . The system of claim 9 , wherein training the mutation machine learning model comprises imparting the mutation machine learning model with a plurality of training sequence of tokens pairs, each training sequence of tokens pair comprising a training input sequence of tokens and a training output sequence of tokens.
11 . The system of claim 9 , wherein the mutation machine learning model comprises a sequence-to-sequence machine learning model.
12 . The system of claim 9 , wherein the token embedding matrix comprises a first dimension associated with an embedding size and a second dimension associated with a quantity of tokens associated with the input sequence of tokens.
13 . The system of claim 9 , wherein the output matrix comprises a first dimension associated with a quantity of possible predictions based at least in part on the numerical index of predefined classification labels and a second dimension associated with a quantity of predictions associated with the output sequence of tokens.
14 . The system of claim 13 , wherein the output sequence of tokens is based at least in part on position values in the first dimension including a highest value for each position in the second dimension.
15 . The system of claim 14 , the one or more output classification labels are based at least in part on integer values in the numerical index of predefined classification labels.
16 . The system of claim 9 , wherein the numerical index of predefined classification labels comprises a set of classification labels mapped to integer values.
17 . One or more non-transitory computer-readable storage media storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
receiving predictive output data from a predictive system, the predictive output data comprising a plurality of classification labels;
generating an input sequence of tokens, the input sequence of tokens comprising a tokenization of the plurality of classification labels based at least in part on a numerical index of predefined classification labels;
generating, using a prediction validation machine learning framework, a mutation of the input sequence of tokens, wherein:
the prediction validation machine learning framework comprises an embedding machine learning model and a mutation machine learning model, wherein:
(i) the embedding machine learning model is configured to (a) generate, for each token of the input sequence of tokens, a token embedding, and (b) concatenate each of the token embedding into a token embedding matrix,
(ii) the mutation machine learning model is configured to generate an output matrix, comprising an output sequence of tokens, based at least in part on the token embedding matrix,
(iii) the mutation of the input sequence of tokens comprises retrieval of one or more output classification labels associated with the output sequence of tokens based at least in part on the numerical index of predefined classification labels; and
performing one or more prediction-based actions based at least in part on the mutation of the input sequence of tokens.
18 . The one or more non-transitory computer-readable storage media of claim 17 , wherein the output matrix comprises a first dimension associated with a number of possible predictions based at least in part on the numerical index of predefined classification labels and a second dimension associated with a quantity of predictions associated with the output sequence of tokens.
19 . The one or more non-transitory computer-readable storage media of claim 18 , wherein the output sequence of tokens is based at least in part on position values in the first dimension including a highest value for each position in the second dimension.
20 . The one or more non-transitory computer-readable storage media of claim 19 , wherein the one or more output classification labels are based at least in part on integer values in the numerical index of predefined classification labels.