Automatic feature learning from a relational database for predictive modelling

Patent metadata

Field	Value
Publication number	US-10762111-B2
Application number	US-201715714140-A
Country	US
Kind code	B2
Filing date	Sep 25, 2017
Priority date	Sep 25, 2017
Publication date	Sep 1, 2020
Grant date	Sep 1, 2020

Abstract

Embodiments for automatic feature learning for predictive modelling in a computing environment by a processor. A first table and a second table are joined based on an edge between the first table and the second table defined by an entity graph thereby creating a resulting joined table that is connected by a column of data. The resulting joined table is used as an input into one or more neural network operations that transform the resulting joined table to one or more features to predict a target variable.

First claim

The invention claimed is:

1. A method, by a processor, for automatic feature learning for predictive modelling in a computing environment, comprising:
joining a first table and a second table based on an edge between the first table and the second table defined by an entity graph, wherein a resulting joined table is connected by a column of data;
using the resulting joined table as an input into one or more neural network operations that transform the resulting joined table to one or more features to predict a target variable;
defining the one or more neural network operations as one or more relational neural network operations and an embedded neural network operation in an embedded layer associated with the resulting joined table; and
inputting numerical data according to a hierarchical order of the resulting joined table into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

2. The method of claim 1 , further including inputting the numerical data of the resulting joined table into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

3. The method of claim 1 , further including:
inputting unstructured data of the resulting joined table into the embedded layer associated with the resulting joined table, wherein unstructured data is from a group consisting of spatial-temporal data, time-series data, sequence data, item set data, number set data, singleton data, text data and image data;
transforming the unstructured data into the numerical data using the embedded neural network operation in the embedded layer associated; and
inputting the numerical data transformed by the embedded neural network operation into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

4. The method of claim 1 , further comprising collecting, by the device, features extracted from the first table and the second table by traversing the entity graph.

5. The method of claim 4 , wherein the entity graph is traversed to a depth based on a defined criterion.

6. A system for automatic feature learning for predictive modelling in a computing environment, comprising:
one or more computers having a physical processor and memory, the memory storing executable instructions that when executed cause the system to:
join a first table and a second table based on an edge between the first table and the second table defined by an entity graph, wherein a resulting joined table is connected by a column of data;
use the resulting joined table as an input into one or more neural network operations that transform the resulting joined table to one or more features to predict a target variable;
define the one or more neural network operations as one or more relational neural network operations and an embedded neural network operation in an embedded layer associated with the resulting joined table; and
input numerical data according to a hierarchical order of the resulting joined table into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

7. The system of claim 6 , wherein the executable instructions that when executed cause the system to input the numerical data of the resulting joined table into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

8. The system of claim 6 , wherein the executable instructions that when executed cause the system to:
input unstructured data of the resulting joined table into the embedded layer associated with the resulting joined table, wherein unstructured data is from a group consisting of spatial-temporal data, time-series data, sequence data, item set data, number set data, singleton data, text data and image data;
transform the unstructured data into the numerical data using the embedded neural network operation in the embedded layer associated; and
input the numerical data transformed by the embedded neural network operation into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

9. The system of claim 6 , wherein the executable instructions that when executed cause the system to collect features extracted from the first table and the second table by traversing the entity graph.

10. The system of claim 9 , wherein the entity graph is traversed to a depth based on a defined criterion.

11. A computer program product for, by a processor, automatic feature learning for predictive modelling in a computing environment, the computer program product comprising a non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions comprising:
an executable portion that joins a first table and a second table based on an edge between the first table and the second table defined by an entity graph, wherein a resulting joined table is connected by a column of data;
an executable portion that uses the resulting joined table as an input into one or more neural network operations that transform the resulting joined table to one or more features to predict a target variable;
an executable portion that defines the one or more neural network operations as one or more relational neural network operations and an embedded neural network operation in an embedded layer associated with the resulting joined table; and
an executable portion that inputs numerical data according to a hierarchical order of the resulting joined table into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

12. The computer program product of claim 11 , further including an executable portion that inputs the numerical data of the resulting joined table into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

13. The computer program product of claim 11 , further including an executable portion that:
inputs unstructured data of the resulting joined table into the embedded layer associated with the resulting joined table, wherein unstructured data is from a group consisting of spatial-temporal data, time-series data, sequence data, item set data, number set data, singleton data, text data and image data;
transforms the unstructured data into the numerical data using the embedded neural network operation in the embedded layer associated; and
inputs the numerical data transformed by the embedded neural network operation into the one or more relational neural network operations that transform the numerical data to one or more features to predict the target variable.

14. The computer program product of claim 11 , further including an executable portion that collects features extracted from the first table and the second table by traversing the entity graph, wherein the entity graph is traversed to a depth based on a defined criterion.

Assignees

• IBM

Inventors

• Buesser Beat
• Hoang Thanh Lam
• Sinn Mathieu
• Tran Ngoc Minh

Classifications

• G06N3/042

  Knowledge-based neural networks; Logical representations of neural networks · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title