CodeTS framework generates time series from text via executable code

Researchers have introduced CodeTS, a novel framework for generating time series data from natural language descriptions. This method reformulates the task into a Text-to-Code-to-Time Series process, where natural language prompts are first translated into executable code that dictates the temporal patterns, and then this code is executed to produce the time series. CodeTS utilizes aligned Text-Code-TS triplets for supervised initialization and employs reinforcement learning with verifiable rewards to train the model without direct code annotations, ensuring format validity, code executability, and time series quality. Experiments across eight benchmarks show that CodeTS outperforms existing LLM-based baselines and supervised generative models, offering a strong zero-shot solution for text-to-time series generation. AI

IMPACT Introduces a novel code-based approach for time series generation, potentially improving accuracy and verifiability over direct LLM generation.

RANK_REASON The cluster contains a research paper detailing a new method for text-to-time series generation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• LLM-based baselines
• Reinforcement Learning with Verifiable Rewards
• Text-to-Code-to-TS
• Text-to-Time Series Generation
• Text-to-TS

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →