9. Normalization Configuration

Normalizations determine how data is transformed before it is used by the machine learning system. Having multiple normalizations will feed multiple values into the system,

 

Each variable needs to have at least 1 normalization, and there are a maximum of 500 normalizations overall.
 
 

TYPES OF NORMALIZATIONS

Normalization Description
Noop Use raw values.
Global Standard Deviation Rescale using the z-score locally on each security over a time window and then rescale globally according the maximum and minimum z-score across all securities.
Local Standard Deviation Rescale using the z-score locally on each security over a time window.
Global N1Top1 Rescale to a scale between negative one and positive across all securities over a time window.
Local N1Top1 Rescale to a scale between negative one and positive for each security locally over a time window.
Global N1Top1 v2 Rescale to a scale between negative one and positive for each security locally over a time window and then rescale to a scale between negative one and positive across all securities.
Global Z1Top1 v2 Rescale to a scale between zero and positive for each security locally over a time window and then rescale to a scale between negative one and positive across all securities.
Global Z1Top1 Rescale to a scale between zero and positive across all securities over a time window.
Local Z1Top1 Rescale to a scale between zero and positive for each security locally over a time window.
Global Standard Deviation Rescale using the z-score locally on each security over an exponentially weighted time window and then rescale globally according the maximum and minimum z-score across all securities.
Local Standard Deviation Rescale using the z-score locally on each security over an exponentially weighted time window.
Global N1Top1 EWM Rescale to a scale between negative one and positive for each security locally over an exponentially weighted time window and then rescale to a scale between negative one and positive across all securities.
Global ZTop1 EWM Rescale to a scale between zero and positive for each security locally over an exponentially weighted time window and then rescale to a scale between negative one and positive across all securities.
Global ZTop1 EWM Rescale to a scale between zero and positive across all securities over an exponentially weighted time window.
Local ZTop1 EWM Rescale to a scale between zero and positive for each security locally over an exponentially weighted time window.
Global N1Top1 EWM Rescale to a scale between negative one and positive across all securities over an exponentially weighted time window.
Local N1Top1 EWM Rescale to a scale between negative one and positive for each security locally over an exponentially weighted time window.