tspy.data_structures.forecasting.ForecastingModel module

class tspy.data_structures.forecasting.ForecastingModel.ForecastingModel(jvm, algorithm, j_fm=None)

Bases: object

An online forecasting model which has the following components: 1. Wraps a forecasting algorithm (see forecasters for algorithms) 2. Updated in real-time with new scalar data, thereby modifying the underlying components of the forecasting algorithm 3. Provides forecasts based on the most recent model updates

Examples

create a forecasting model (auto-forecasting algorithm)

>>> import tspy
>>> model = tspy.forecasters.auto(10)
>>> model
Forecasting Model
  Algorithm: Unitialized.  Min training size=10
HWSAdditive:HWSAdditive=-1 (aLevel=0, bSlope=0, gSeas=0) uninitialized
HWSMultiplicative:HWSMultiplicative=-1 (aLevel=0, bSlope=0, gSeas=0) uninitialized
RegularARIMAAlgorithm:RegularARIMAAlgorithm [armaAlgorithm=RegularARMAAlgorithm [dataHistory=null, errorHistory=null, dataMean=NaN, minTrainingData=80, pMin=0, pMax=5, qMin=0, qMax=5, forceModel=false], differencer=null, minTrainingData=82, diffOrder=-1]
RegularSegmentedLinear:RegularSegmentedLinear,history=-1,width=1 [ ]
RegularSegmentedAveraging:RegularSegmentedAveraging,history=-1,width=1 [uninitialized ]
RegularLinear:RegularLinear [linearFitter=LeastSquaresFitter [yIntercept=0.0, slope=0.0], minTrainingSamples=2]
RegularLinear:RegularLinear [linearFitter=LeastSquaresFitter [yIntercept=0.0, slope=0.0], minTrainingSamples=2]
BATSAlgorithm:BATSAlgorithm

update the model with values

>>> for i in range(0,10):
    ...model.update_model(i, float(i))
>>> model
Forecasting Model
  Algorithm: HWSAdditive:HWSAdditive=1 (aLevel=0.001, bSlope=0.001, gSeas=0.001) level=9.99004488020975, slope=0.9999900896408388, seasonal(amp,per,avg)=(0.0,1, 5,-0.009900448802097483)
HWSMultiplicative:HWSMultiplicative=1 (aLevel=0.001, bSlope=0.001, gSeas=0.001) level=9.990109117210386, slope=0.9999901537464366, seasonal(amp,per,avg)=(0.0,1, 5,0.9970972695898375)
RegularARIMAAlgorithm:RegularARIMAAlgorithm [armaAlgorithm=RegularARMAAlgorithm [dataHistory=null, errorHistory=null, dataMean=NaN, minTrainingData=80, pMin=0, pMax=5, qMin=0, qMax=5, forceModel=false], differencer=null, minTrainingData=82, diffOrder=-1]
RegularSegmentedLinear:RegularSegmentedLinear=1,history=-1,width=1 [(y=1.0*x + 0.0)  ]
RegularSegmentedAveraging:RegularSegmentedAveraging=1,history=-1,width=1 [(avg=4.5)  ]
RegularLinear:RegularLinear [linearFitter=LeastSquaresFitter [yIntercept=0.0, slope=1.0], minTrainingSamples=2]
RegularLinear:RegularLinear [linearFitter=LeastSquaresFitter [yIntercept=0.0, slope=1.0], minTrainingSamples=2]
BATSAlgorithm:BATSAlgorithm

forecast at a time in the future

>>> model.forecast_at(15)
14.993695899928097

Attributes

average_intervalfloat

Returns ——- float Get the average interval across all calls to update_model().

error_horizon_lengthint

Returns ——- int Get the value of the error horizon length.

initial_time_updatedint

Returns ——- int Get the first time value used to build the forecasting model.

last_time_updatedint

Returns ——- int Get the most recent time value used to build the forecasting model.

Methods

bounds_at(at, confidence)	get the lower and upper bounds for the forecast at the given time and the given confidence level
error_at(at)	compute the error estimate for the forecast at the given point in time
forecast_at(at)	compute the forecast at a given time
is_initialized()	Returns
reset_model()	reset the model to an uninitialized state
save(path)	save the model to a file
update_model(timestamp, value)	update the model at a given timestamp

property average_interval

Returns

float

Get the average interval across all calls to update_model(). Will return -1 if 1 or fewer samples have been provided.

bounds_at(at, confidence)

get the lower and upper bounds for the forecast at the given time and the given confidence level

Parameters

atint

the time in the future, for which we want to predict the bounds with the confidence value that corresponds to the forecast at that time

confidencefloat

a number between 0 and 1 representing the confidence required for the return bounds values

Returns

list

a list of size 2 containing the lower and upper bounds, in this order, such that the actual value is between these bounds with probability equal to the given confidence. Never return null, however, if the bounds could not be computed, then the lower bound will be negative infinity (min float) and the upper bound will be positive infinity (max float)

error_at(at)

compute the error estimate for the forecast at the given point in time

Parameters

atint

time in the future at which error is being requested

Returns

float

the error estimate. If the underlying algorithm does not support error computation, returns NaN

property error_horizon_length

Returns

int

Get the value of the error horizon length. Will return 0 if not set.

forecast_at(at)

compute the forecast at a given time

Parameters

atint

the time at which to compute a forecast

Returns

float

a valid forecast value

property initial_time_updated

Returns

int

Get the first time value used to build the forecasting model. Will return a negative value if the model has not been initialized

is_initialized()

Returns

bool

True if the model has been initialized and is ready for forecasting

property last_time_updated

Returns

int

Get the most recent time value used to build the forecasting model. Will return negative value if the model is not updated

reset_model()

reset the model to an uninitialized state

save(path)

save the model to a file

Parameters

pathstring

path to file

update_model(timestamp, value)

update the model at a given timestamp

Parameters

timestampint

time corresponding to the given data value

valuefloat

data value to be trained into the model at the given timestamp