Machine Learning with window data

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Machine Learning with window data

chris-sw

Hi,

 

I have a use case where I like to analyze windows of sensordata.

Currently I have a working case where I use Structured Streaming to process real-time streams of sensordata. Now I like to analyse windows of sensordata and use classification to predict the class of a whole window.

For instance, the application receives batches of sensordata (where each record holds: timestamp, value, key). With the use of Machine learning I like to analyse windows of these streams and classify the window as ‘warm’ or ‘cold’. A single record is not sufficient for classification, a window of records shapes a pattern to be used for classification.

 

But how should you define features for a window of sensordata?

Each value (sensor) as a separate feature in the vector (for a window of x seconds, the vector contains x sensor values)? Or is there a way a feature can hold multiple values (like an array)? Or use some kind of encoding to fit x sensor values as a single feature?

 

Regards,

Chris

 

Reply | Threaded
Open this post in threaded view
|

Re: Machine Learning with window data

Robb Greathouse
I keep unsubscribing from this list; but continue to receive emails.

On Fri, Aug 3, 2018 at 4:01 AM Christiaan Ras <[hidden email]> wrote:

Hi,

 

I have a use case where I like to analyze windows of sensordata.

Currently I have a working case where I use Structured Streaming to process real-time streams of sensordata. Now I like to analyse windows of sensordata and use classification to predict the class of a whole window.

For instance, the application receives batches of sensordata (where each record holds: timestamp, value, key). With the use of Machine learning I like to analyse windows of these streams and classify the window as ‘warm’ or ‘cold’. A single record is not sufficient for classification, a window of records shapes a pattern to be used for classification.

 

But how should you define features for a window of sensordata?

Each value (sensor) as a separate feature in the vector (for a window of x seconds, the vector contains x sensor values)? Or is there a way a feature can hold multiple values (like an array)? Or use some kind of encoding to fit x sensor values as a single feature?

 

Regards,

Chris

 



--
Robb Greathouse
Middleware BU
505-507-4906