Introduction

Introduction to TIM

TIM, or Tangent Information Modeler, is Tangent Works’ automatic model building engine. It is designed specifically for time series forecasting and anomaly detection.

It has been tested numerous times. TIM is proven to generate high-quality models with zero degrees of freedom, meaning no tuning of engine parameters is required from a user. Impressively, TIM in business user mode won the General Energy Forecasting Competition (GEFCom) in 2017, where it built over 150 models.

InstantML

TIM differs to Automatic Machine Learning (AutoML) strategies that have recently drawn the attention of many researchers and industry professionals.

AutoML focusses on the selection of an appropriate modeling technique and its hyperparameters for a task at hand. In this effort, AutoML solutions scan through many different ML libraries, create models and tune their corresponding hyperparameters. This has traditionally been a laborious task.

In time series modeling however, we believe that identification of significant features and the overall modeling framework (how to address changing dynamics in a time series, dynamic data availability, multi-situational forecasts, etc.) are far more important than the choice of a specific modeling technique and its associated hyperparameters.

And so, we created InstantML, or Instant Machine Learning for time series data. With InstantML TIM generates one single high-quality model with a single pass through the data. It is a modeling strategy that identify relevant features present in the data. And, it works significantly faster (seconds up to minutes on standard hardware) than typical AutoML strategies.

RTInstantML

Reducing steps required in forecasting process, especially in large scale operations, is possible due to advancements in InstantML. We took forecasting to the next level and unlocked use of machine learning technology to work almost in real time. How does it work?

TIM looks in the latest data and detects a corresponding data availability and forecasting routine out of it. There is no need to setup your data availability scheme for the situation – it is defined by the way data are organized and TIM will recognize this automatically.

TIM will then generate a model and produce a forecast in one single request. The model is then discarded, as the next time a forecast is required the entire process is simply repeated. This is inherently different to AutoML. All TIM needs to know from a user is how many steps ahead the forecast should be calculated.

It is especially advantageous if data availability scheme changes often or if forecasts are required ad-hoc, i.e. at any time of a day.

RTInstantML does not require any model management nor model storage, which simplifies implementation complexity and shortens time to value.