my_ml_notes

Forecasting

Timeseries 101

Univariate: Only one variable is varying over time. For example, data collected from a sensor measuring the temperature of a room every second. Therefore, each second, you will only have a one-dimensional value, which is the temperature.

Multivariate: Multiple variables are varying over time. For example, a tri-axial accelerometer. There are three accelerations, one for each axis (x,y,z) and they vary simultaneously over time.

Deep Learing-based Forecasting or Neural Forecasting

Leverages deep neural networks to predict time series into the future. It requires less feature engineering compared to classical models such as ARIMA and ETS.

Problem: Cold start problem

Issue: Forecast articles that have little or no historical sales data.

Idea: learn typical behaviour of new articles based on patterns of other types of articles.

Types of neural forecast:

• Local models: trained for each time series.
• Global models: trained accross multiple related time series.It can address cold start problem as it uses learned patterns, also known as embeddings, to produce forecasts.

Models:

DeepAR: autoregressive model based on RNN (recurrrent neural network) for univariate timeseries. DeepAR produces a probabilistic forecast (i.e. a set of possible outcomes, each with a certain probability).

References:

• Blog: How adidas and AWS optimised supply chain management to satisfy customer demand through Machine Learning and MLOps on Amazon SageMaker

This site is open source. Improve this page.