Embedding models for time series data [D]
Fourier-domain embeddings could unlock flexible feature extraction for time series data.
Deep Dive
A Reddit user is asking the community for open source embedding models that work on time series data, specifically ones that operate in the frequency domain via Fourier transforms to handle variable-length sequences.
Key Points
- Request targets open source models for time series embeddings, not proprietary solutions.
- Fourier transform usage would enable support for variable-length series by operating on the frequency domain.
- Current gaps mean most time series feature engineering is manual; embeddings could automate and standardize it.
Why It Matters
Standardized time series embeddings would accelerate work in finance, healthcare, and IoT by enabling transfer learning and reducing manual feature engineering.