Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PromptCast: A New Prompt-based Learning Paradigm for Time Series Forecasting

Hao Xue, Flora D. Salim

TL;DR

PromptCast reframes time series forecasting as a language-model task by transforming numerical inputs and targets into prompts, enabling sentence-to-sentence forecasting with foundation models. The authors release PISA, a large-scale dataset with 311,932 instances across three real-world domains, using a 15-day observation window ($t_{ ext{obs}}=15$) and a 7:1:2 train/val/test split to benchmark both numerical methods and language models. Across extensive experiments, language-model-based PromptCast achieves competitive RMSE/MAE and exhibits strong zero-shot generalization, while ChatGPT-like APIs show higher errors and costs. The work highlights the potential for code-less forecasting, prompts-based experimentation, and future directions including multi-step, multivariate, auto-prompting, and explainable PromptCast applications.

Abstract

This paper presents a new perspective on time series forecasting. In existing time series forecasting methods, the models take a sequence of numerical values as input and yield numerical values as output. The existing SOTA models are largely based on the Transformer architecture, modified with multiple encoding mechanisms to incorporate the context and semantics around the historical data. Inspired by the successes of pre-trained language foundation models, we pose a question about whether these models can also be adapted to solve time-series forecasting. Thus, we propose a new forecasting paradigm: prompt-based time series forecasting (PromptCast). In this novel task, the numerical input and output are transformed into prompts and the forecasting task is framed in a sentence-to-sentence manner, making it possible to directly apply language models for forecasting purposes. To support and facilitate the research of this task, we also present a large-scale dataset (PISA) that includes three real-world forecasting scenarios. We evaluate different SOTA numerical-based forecasting methods and language generation models. The benchmark results with various forecasting settings demonstrate the proposed PromptCast with language generation models is a promising research direction. Additionally, in comparison to conventional numerical-based forecasting, PromptCast shows a much better generalization ability under the zero-shot setting.

PromptCast: A New Prompt-based Learning Paradigm for Time Series Forecasting

TL;DR

PromptCast reframes time series forecasting as a language-model task by transforming numerical inputs and targets into prompts, enabling sentence-to-sentence forecasting with foundation models. The authors release PISA, a large-scale dataset with 311,932 instances across three real-world domains, using a 15-day observation window () and a 7:1:2 train/val/test split to benchmark both numerical methods and language models. Across extensive experiments, language-model-based PromptCast achieves competitive RMSE/MAE and exhibits strong zero-shot generalization, while ChatGPT-like APIs show higher errors and costs. The work highlights the potential for code-less forecasting, prompts-based experimentation, and future directions including multi-step, multivariate, auto-prompting, and explainable PromptCast applications.

Abstract

This paper presents a new perspective on time series forecasting. In existing time series forecasting methods, the models take a sequence of numerical values as input and yield numerical values as output. The existing SOTA models are largely based on the Transformer architecture, modified with multiple encoding mechanisms to incorporate the context and semantics around the historical data. Inspired by the successes of pre-trained language foundation models, we pose a question about whether these models can also be adapted to solve time-series forecasting. Thus, we propose a new forecasting paradigm: prompt-based time series forecasting (PromptCast). In this novel task, the numerical input and output are transformed into prompts and the forecasting task is framed in a sentence-to-sentence manner, making it possible to directly apply language models for forecasting purposes. To support and facilitate the research of this task, we also present a large-scale dataset (PISA) that includes three real-world forecasting scenarios. We evaluate different SOTA numerical-based forecasting methods and language generation models. The benchmark results with various forecasting settings demonstrate the proposed PromptCast with language generation models is a promising research direction. Additionally, in comparison to conventional numerical-based forecasting, PromptCast shows a much better generalization ability under the zero-shot setting.
Paper Structure (24 sections, 4 figures, 14 tables)

This paper contains 24 sections, 4 figures, 14 tables.

Table of Contents

  1. Introduction
  2. Prompt-Based Time Series Forecasting
  3. Dataset Design and Description
  4. Data Sources and Processing
  5. Template-Based Prompting
  6. Statistics Overview
  7. Benchmark
  8. Evaluation Metrics
  9. Baselines
  10. Implementation Details
  11. Experimental Performance
  12. Numerical-Based Methods
  13. Pre-trained Language Models
  14. Performance of ChatGPT
  15. Computation Cost
  16. ...and 9 more sections

Figures (4)

  • Figure 1: Conceptual illustrations of: (a) the typical framework of existing numerical-based forecasting method; (b) the framework of the proposed PromptCast; (c) a potential forecasting chatbot application based on PromptCast.
  • Figure 2: The distribution plots of three sub-sets: (a) the City Temperature (CT) sub-set includes negative values; (b) the Electricity Consumption Load (ECL) sub-set covers large numbers spanning a wide range; and (c) the sub-set of SafeGraph Human Mobility Data (SG) involves relatively small values ranging from 0 to 400.
  • Figure 3: Visualizations of attentions in the proposed PromptCast.
  • Figure 4: The plots of cost and RMSE accuracy on three sub-sets.