Skip to content

Overview

There are two ways how to create a forecast with TIM:

  • The InstantML approach is comprised of model building as a first step and then using the model to create a forecast as a second step. Once the model is built, it can be used for forecasting repeatedly.
  • The RTInstantML approach builds a model and creates a forecast in a one-step process. The process is repeated every time a new forecast is desired, thus, each model is used for forecasting only once and then discarded.

The following sections serve as an overview of the structure of the model building and forecasting tasks, the inputs as well as the outputs.

TIM InstantML Forecasting structure

The image below shows the two-step process of TIM InstantML Forecasting - the model building phase on the left and the forecasting phase on the right.

build_model_and_predict

In order to create a model, TIM requires the following inputs:

  • Data - Historical data used to build the model. Read more about the required properties in the Data properties section.

  • Data updates - A cron-like notation that describes the routine of how the target variable and predictors are updated. This information together with the usage input (contained in the input configuration) define the scenario for which the model is built.

  • Configuration - Math settings and other configuration parameters.

These inputs are processed by the TIM engine during the model building phase. The following is returned from the model building phase of TIM:

  • Model - Returned as an xml file.

  • Outputs - Predictions, predictor importances, etc.

The model can be then used for forecasting. In order to create a forecast new data have to be provided and optionally a prediction scope configuration can be specified. The outputs from the forecasting phase are described in the same Outputs section as discussed in the model building phase above.

TIM RTInstantML Forecasting structure

Below is a sketch of the TIM RTInstantML Forecasting approach.

build_model_predict

The following inputs are required to create an RTInstantML forecast:

  • Data - Historical data used to build a model. Read more about the required properties in the Data properties section.

  • Configuration - Math settings and other configuration parameters.

These inputs are processed by TIM engine in the model building phase followed by a forecasting phase where the desired forecast is created. The resulting predictions and other outputs are described in the Outputs section.