FeatureTimeSeries

Overview

Time series options for a feature.

type:

object

Properties

delta_max

If specified, ensures that the largest difference between feature values is not larger than this specified value. A null value means no max boundary. The length of the list must match the number of derivatives as specified by order. Only applicable when time series type is set to delta.

type:

list

values:

number

delta_min

If specified, ensures that the smallest difference between features values is not smaller than this specified value. A null value means no min boundary. The length of the list must match the number of derivatives as specified by order. Only applicable when time series type is set to delta.

type:

list

values:

number

derived_orders

The number of orders of derivatives that should be derived instead of synthesized. Ignored if order is not provided.

type:

number

lags

If specified, generates lag features containing previous values using the enumerated lag offsets. Takes precedence over num_lags. If neither num_lags nor lags is specified for a feature, then a single lag feature is generated.

type:

list

values:

number

num_lags

If specified, generates the specified amount of lag features containing previous values. If lags is specified, then this parameter will be ignored. If neither num_lags nor lags is specified for a feature, then a single lag feature is generated.

type:

number

order

If provided, will generate the specified number of derivatives and boundary values.

type:

number

rate_max

If specified, ensures that the rate (the difference quotient, the discrete version of derivative) for this feature won’t be more than the value provided. A null value means no max boundary. The value must be in epoch format for the time feature. The length of the list must match the number of derivatives as specified by order. Only applicable when time series type is set to rate.

type:

list

values:

number

rate_min

If specified, ensures that the rate (the difference quotient, the discrete version of derivative) for this feature won’t be less than the value provided. A null value means no min boundary. The value must be in epoch format for the time feature. The length of the list must match the number of derivatives as specified by order. Only applicable when time series type is set to rate.

type:

list

values:

number

time_feature

When true, the feature will be treated as the time feature for time series modeling. Additionally, time features must use type delta.

type:

boolean

type

When rate is specified, uses the difference of the current value from its previous value divided by the change in time since the previous value. When delta is specified, uses the difference of the current value from its previous value regardless of the elapsed time. Set to delta if feature has time_feature set to true. When covariate is specified, temporal changes are not modeled and feature values are directly predicted with interpolation in series generation rather than derived using a rate or delta.

type:

string

enum:

rate | delta | covariate

universal

Controls whether future values of independent time series are considered. Applicable only to the time feature. When false, the time feature is not universal and allows using future data from other series in decisions; this is applicable when the time is not globally relevant and is independent for each time series. When true, universally excludes using any data with from the future from all series; this is applicable when time is globally relevant and there are events that may affect all time series. If there is any possibility of global relevancy of time, it is generally recommended to set this value to true, which is the default.

type:

boolean