Feature processing tradeoff management
AssigneeAmazon Technologies, Inc.
InventorsLeo Parker Dirac, Nicolle M. Correa, Charles Eric Dannaker
Life of this patent
Estimating the term…
Abstract
At a machine learning service, a set of candidate variables that can be used to train a model is identified, including at least one processed variable produced by a feature processing transformation. A cost estimate indicative of an effect of implementing the feature processing transformation on a performance metric associated with a prediction goal of the model is determined. Based at least in part on the cost estimate, a feature processing proposal that excludes the feature processing transformation is implemented.
Claims20
A system, comprising:
- one or more computing devices configured to:
- receive, from a client of a machine learning service, (a) a training data set to be used to train a machine learning model, (b) a model quality metric for evaluating the machine learning model, and (c) an indication of a recipe that specifies one or more feature processing transformations to be performed on respective input variables in training data set;
- identify a plurality of candidate feature processing transformations of the input variables to be included in the recipe;
- determine a quality estimate indicative of an effect, on the model quality metric, of including a particular one of the candidate feature processing transformations in training the machine learning model;
- and
- provide, to the client and based at least in part on the quality estimate, a recommendation to include or exclude the particular candidate feature processing transformation in the recipe.
A computer-implemented method, comprising:
- performing, by one or more computing devices in a machine learning service:
- receiving, from a client of the machine learning service, (a) a training data set to be used to train a machine learning model, (b) a model quality metric for evaluating the machine learning model, and (c) an indication of a recipe that specifies one or more feature processing transformations to be performed on respective input variables in training data set;
- identifying a plurality of candidate feature processing transformations of the input variables to be included in the recipe;
- determining a quality estimate indicative of an effect, on the model quality metric, of including a particular one of the candidate feature processing transformations in training the machine learning model;
- and
- providing, to the client and based at least in part on the quality estimate, a recommendation to include or exclude the particular candidate feature processing transformation in the recipe.
One or more non-transitory computer-accessible storage media storing program instructions that when executed on or across one or more processors of a machine learning service, cause the machine learning service to:
- receive, from a client of the machine learning service, (a) a training data set to be used to train a machine learning model, (b) a model quality metric for evaluating the machine learning model, and (c) an indication of a recipe that specifies one or more feature processing transformations to be performed on respective input variables in training data set;
- identify a plurality of candidate feature processing transformations of the input variables to be included in the recipe;
- determine a quality estimate indicative of an effect, on the model quality metric, of including a particular one of the candidate feature processing transformations in training the machine learning model;
- and
- provide, to the client and based at least in part on the quality estimate, a recommendation to include or exclude the particular candidate feature processing transformation in the recipe.
Dependencies are inferred from the claim text.