# US 2021/0241171 A1 · Machine learning feature engineering

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0241171 A1
- Document ID: US20210241171A1
- Status: Published application
- Kind: A1
- Application number: 16/877,407
- Filed: 2020-05-18
- Published: 2021-08-05
- Earliest claimed: 2020-02-03
- Assignees: Kaskada, Inc.
- Applicants: Kaskada, Inc.
- Inventors: Davor Bonaci; Benjamin Chambers; Andrew Concordia; Emily Kruger; Ryan Michael
- Web page: https://patentgrep.com/patent/US20210241171A1

## CPC

- G06N 5/022: Knowledge engineering; Knowledge acquisition
- G06N 20/00: Machine learning
- G06F 16/284: {Relational databases}
- G06N 5/02: Knowledge representation; Symbolic representation
- G06N 5/04: Inference or reasoning models
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Machine learning feature engineering systems and methods comprise an event ingestion module that receives event data associated with entities. The ingestion module determines which entities are associated with events of the event data. The ingestion module stores the events, grouped by associated entity, in a related event store. A user defines features associated with the entities via an API and/or a feature studio. A feature computation layer determines values for the features based on the grouped events stored to the related event store. The feature computation layer stores the computed feature values and timestamps to a feature store. When new data is received, the feature computation layer computes one or more of the feature values for different times based on the timestamps. Feature vectors are generated using the computed feature values and output to the user via the API and/or feature studio.

## Claims

### Claim 1

A method comprising:
receiving, from a data source by a feature engine, a data stream indicating a plurality of events associated with a plurality of entities;
generating, by the feature engine, an event dataset comprising groups of the plurality of events associated with one or more entities of the plurality of entities wherein the event dataset comprises a plurality of timestamps associated with the plurality of events, wherein each of the plurality of timestamps comprises a time component;
receiving, an indication of one or more features associated with the plurality of entities and a configuration for determining the one or more features;
generating, using the configuration and the event dataset, a plurality of feature vectors comprising the one or more features;
receiving additional event data;
determining, based on the plurality of timestamps, to re-generate at least some of the the one or more features in the feature vectors based on the additional event data;
re-generating a feature vector using at least one of the generated plurality of feature vectors, wherein the generated plurality of feature vectors comprise lists of values of the one or more features and wherein re-generating the feature vector comprises updating at least some of the lists of values using the additional event data; and
outputting the re-generated feature vector associated with at least one entity.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
determining an anchor time; and
computing the values of the one or more features based on events that occurred prior to the anchor time, wherein the events are retrieved from historical data.

### Claim 3 (depends on 1)

The method of claim 1, wherein each of the plurality of timestamps comprises:
a unique machine identifier; and
a sequence number.

### Claim 4 (depends on 1)

The method of claim 1, further comprising:
determining, based on the plurality of timestamps, to replace at least a portion of the event dataset using the additional event data.

### Claim 5 (depends on 1)

The method of claim 1, wherein the feature vector dataset comprises a plurality of timestamps associated with the plurality of feature vectors, and wherein the method further comprises:
determining, based on the plurality of timestamps, a portion of the feature values to compute for different times.

### Claim 6 (depends on 1)

The method of claim 1, wherein the data indicating the plurality of events comprises a stream of data indicating at least one of live events, events that occurred after a transmission delay, or historical events that were republished to the stream.

### Claim 7 (depends on 1)

The method of claim 1, wherein the method further comprises storing definitions of the one or more features;
and wherein generating the feature vector dataset comprises using the stored definitions.

### Claim 8

A non-transitory computer-readable medium storing instructions that, when executed, cause operations comprising:
receiving, from one or more data sources by a feature engine, a data stream indicating a plurality of events associated with a plurality of entities wherein the event dataset comprises a plurality of timestamps associated with the plurality of events, wherein each of the plurality of timestamps comprises a time component;
generating, by the feature engine, an event dataset comprising groups of the plurality of events associated with one or more entities of the plurality of entities;
receiving, via an application programming interface (API), an indication of one or more features associated with the plurality of entities and a configuration for determining the one or more features;
generating, using the configuration and the event dataset, a feature vector dataset comprising a plurality of feature vectors associated with the plurality of entities, wherein the plurality of feature vectors comprise lists of values of the one or more features;
receiving additional event data in the data stream indicating an additional plurality of events;
generating, based on the plurality of timestamps, at least one of the values of the plurality of feature vectors in the generated feature vector dataset based on updating the lists of values using the list of values and the additional event data;
receiving, via the API, an indication of at least one entity of the plurality of entities; and
outputting at least one feature vector, from the feature vector dataset, associated with the at least one entity.

### Claim 9 (depends on 8)

The non-transitory computer-readable medium of claim 8, wherein the operations further comprise:
determining an anchor time; and
computing the values of the one or more features based on events that occurred prior to the anchor time, wherein the events are retrieved from historical data.

### Claim 10 (depends on 8)

The non-transitory computer-readable medium of claim 8, wherein each of the plurality of timestamps comprises:
a unique machine identifier; and
a sequence number.

### Claim 11 (depends on 8)

The non-transitory computer-readable medium of claim 8, wherein the operations further comprise:
determining, based on the plurality of timestamps, to replace at least a portion of the event dataset using the additional event data.

### Claim 12 (depends on 8)

The non-transitory computer-readable medium of claim 8, wherein the feature vector dataset comprises a plurality of timestamps associated with the plurality of feature vectors, and wherein the operations further comprise:
determining, based on the plurality of timestamps, a portion of the feature values to compute for different times.

### Claim 13 (depends on 8)

The non-transitory computer-readable medium of claim 8, wherein the data indicating the plurality of events comprises a stream of data indicating at least one of live events, events that occurred after a transmission delay, or historical events that were republished to the stream.

### Claim 14 (depends on 8)

The non-transitory computer-readable medium of claim 8, wherein the data indicating the plurality of events comprises at least one of stored historical events or one or more streams of data.

### Claim 15

A feature engine comprising:
an event store configured to store an event dataset comprising groups of a plurality of events associated with one or more entities of a plurality of entities;
at least one processor;
a non-transitory computer-readable memory coupled to the at least one processor, the computer-readable memory having stored thereon computer readable instructions that upon execution on the at least one processor at least cause the feature engine to:
receive, from a data source, a stream of data indicating the plurality of events associated with the plurality of entities wherein the plurality of events comprise a timestamp associated with each of the plurality of events, wherein the timestamps comprises a time component;
generate, using the event dataset, a feature vector dataset comprising a plurality of feature vectors associated with the plurality of entities, wherein the plurality of feature vectors comprise lists of values of one or more features;
receive an indication of an additional plurality of events in the stream of data;
re-generate, based on the timestamps, at least one of the values in the plurality of feature vectors in the generated feature vector dataset based on updating the lists of values using the list of values and the additional plurality of events;
receive an indication of at least one entity of the plurality of entities; and
output at least one feature vector, from the feature vector dataset, associated with the at least one entity.

### Claim 16 (depends on 15)

The feature engine of claim 15, wherein the feature engine is further configured to:
determine an anchor time; and
compute the values of the one or more features based on events that occurred prior to the anchor time, wherein the events are retrieved from historical data.

### Claim 17 (depends on 15)

The feature engine of claim 15, wherein the event dataset comprises a plurality of timestamps associated with the plurality of events, wherein each of the plurality of timestamps comprises:
a unique machine identifier; and
a sequence number.

### Claim 18 (depends on 15)

The feature engine of claim 15, wherein the feature engine is further configured to:
determine, based on the plurality of timestamps, to replace at least a portion of the event dataset using the additional event data.

### Claim 19 (depends on 17)

The feature engine of claim 17, wherein the feature engine is further configured to:
determining, based on the plurality of timestamps, a portion of the feature values to compute for different times.

### Claim 20 (depends on 15)

The feature engine of claim 15, wherein the data indicating the plurality of events comprises at least one of a stream of data indicating at least one of live events, events that occurred after a transmission delay, or historical events that were republished to the stream.
