Einige Inhalte dieser Anwendung sind momentan nicht verfügbar.
Wenn diese Situation weiterhin besteht, kontaktieren Sie uns bitte unterFeedback&Kontakt
1. (WO2019005049) ITERATIVE FEATURE SELECTION METHODS
Anmerkung: Text basiert auf automatischer optischer Zeichenerkennung (OCR). Verwenden Sie bitte aus rechtlichen Gründen die PDF-Version.

CLAIMS

What is claimed is:

1 . A method of decreasing computation time required to improve models that relate predictors and outcomes in a dataset, the method comprising:

generating a first model, wherein the first model comprises a first model component from a first pool of model components;

generating a second model, wherein the second model comprises a second model component from a second pool of model components;

computing a first utility metric of the first model component comprising a ratio of (1 ) a quantity of models in which the first model component is present, to (2) a quantity of model component pools in which the first model component is present;

computing a second utility metric of the second model component comprising a ratio of (1 ) a quantity of models in which the second model component is present, to (2) a quantity of model component pools in which the second model component is present; and

eliminating, based on the first and second utility metrics, the first and second model components from the first and second pools of model components.

2. The method of claim 1 , wherein the first model component is randomly

generated.

3. The method of claim 1 , wherein the first and second pools of model components comprise at least one of: a computational operator, a mathematical operator, a constant, a predictor, a feature, a variable, a ternary operator, an algorithm, a formula, a binary operator, a hidden node, a weight, a bias, a gradient, a hyper- parameter.

4. The method of claim 1 , wherein the first function comprises a product of at least the first model-attribute and the first utility metric.

5. The method of claim 4, wherein the first function and the second function are the same.

6. The method of claim 1 , wherein the steps of generating a first and second model comprises an iterative modeling process.

7. The method of claim 6, wherein the iterative modeling process comprises at least one of: an evolutionary computing process, a genetic programming process, a genetic algorithm process, a neural network process, a deep learning process, a Markov modeling process, a Monte Carlo modeling process, and a stepwise regression process.

8. The method of claim 1 , further comprising the step of retaining, based on the first and second utility metrics, the first and second model components from the first and second pools of model components.

9. The method of claim 1 , further comprising the steps of:

eliminating, based on the first and second utility metrics, the first and second model components from the first and second pools of model components to generate a third pool of model components; generating a third model, wherein the third model comprises a third model component from the third pool of model components;

computing a third utility metric of the third model component comprising a ratio of (1 ) a quantity of models in which the third model component is present, to (2) a quantity of model component pools in which the third model component is present; and

eliminating, based on the third utility metric, the third model component from the third pool of model components.

10. A method of decreasing computation time required to improve models that relate predictors and outcomes in a dataset, the method comprising:

generating a first model, wherein the first models comprises a first model component from a first pool of model components;

generating a second model, wherein the second model comprises a second model component from a second pool of model components;

computing (1 ) a first model-attribute metric corresponding to the first model and (2) a second model-attribute metric corresponding to the second model;

computing a first utility metric of the first model component comprising a ratio of (1 ) a quantity of models in which the first model component is present, to (2) a quantity of model component pools in which the first model component is present;

computing a second utility metric of the second model component comprising a ratio of (1 ) a quantity of models in which the second model component is present, to (2) a quantity of model component pools in which the second model component is present;

computing a first weighted utility metric that corresponds to the first model component, the first weighted utility metric comprising an outcome of a first function that incorporates: (1 ) model-attribute metrics for models in which the first model component is present and (2) the first utility metric;

computing a second weighted utility metric that corresponds to the second model component, the weighted utility metric comprising an outcome of a second function comprising: (1 ) model-attribute metrics for models in which the second model component is present and (2) the second utility metric; and

eliminating, based on the first and second weighted utility metrics, the first and second model components from the first and second pools of model components.

1 1 . The method of claim 10, wherein the model-attribute metric comprises at least one of accuracy, sensitivity, specificity, area under curve (AUC) from a receiver operating characteristic (ROC) metric, and algorithm length.

12. The method of claim 10, wherein the steps of generating a first and second

model comprises an iterative modeling process.

13. The method of claim 12, wherein the iterative modeling process comprises at least one of: an evolutionary computing process, a genetic programming process, a genetic algorithm process, a neural network process, a deep learning process, a Markov modeling process, a Monte Carlo modeling process, and a stepwise regression process.

14. A method of decreasing computation time required to improve models that relate predictors and outcomes in a dataset, the method comprising:

generating a model comprising a model component;

computing, using a subset of the dataset, a model-attribute metric

corresponding to the model;

computing a utility metric of the model component comprising a ratio, wherein a numerator of the ratio comprises a quantity of models in which the model component is present;

wherein a denominator of the ratio begins at zero and is incremented by one when the model component is present in a pool of model components; computing a weighted utility metric that corresponds to the model component, the weighted utility metric comprising an outcome of a function that incorporates: (1 ) the model-attribute metric and (2) the utility metric; and

eliminating, based on the weighted utility metric, the model component from the pool of model components.

15. The method of claim 14, further comprising the step of retaining the model

component from the pool of model components based on the weighted utility metric.

16. The method of claim 14, wherein the model component is randomly generated.

17. The method of claim 14, wherein the model component comprises at least one of a computational operator, a mathematical operator, a constant, a predictor, a feature, a variable, a ternary operator, an algorithm, a formula, a binary operator, a hidden node, a weight, a bias, a gradient, a hyper-parameter.

18. The method of claim 14, wherein the pool of model components comprises at least one of: a computational operator, a mathematical operator, a constant, a predictor, a feature, a variable, a ternary operator, an algorithm, a formula, and a binary operator.

19. The method of claim 14, wherein the function comprises a product of at least the model-attribute and the utility metric.

0. The method of claim 14, wherein the model-attribute comprises at least one of accuracy, sensitivity, specificity, area under curve (AUC) from a receiver operating characteristic (ROC) metric, and algorithm length.