Patch

As explained in Time Series Forecasting Using Foundation Models

A patch is a segment of consecutive time-series observations grouped as one model input unit.

A patch groups consecutive time-series observations into one token or model input unit. Grouping local points helps models learn nearby relationships while reducing the number of tokens sent to the embedding layer. section 6.1.1

Patching supports several implementation choices:

• A patch length of two groups two observations before tokenization, rather than sending each observation separately. section 6.1.1
• Moirai selects patch lengths according to data frequency, while Moirai-MoE uses one fixed length and nonoverlapping patches across series. section 6.2.1 section 6.1.1
• Time-LLM normalizes and patches the series before embedding patches and translating them through learned text prototypes. section 9.1.1

Patch reprogramming example

A patch can receive descriptions such as “down steady” or “up short”; combining prototypes can describe a patch as “down steady then up short.” section 9.1.1

Figure 6.2 The process of patching. Typically, each data point (circle) is a token that’s sent to the embedding layer. With patching, we group data points, and each patch becomes a token that is sent to the embedding layer.

Shows how multiple time-series points become one token before entering the embedding layer. section 6.1.1

How Time Series Forecasting Using Foundation Models covers it

How deep it goes

Explains grouping, tokenization, embedding, token reduction, and frequency-dependent lengths, without presenting implementation code.

Examples

Demonstrates length-two patches, frequency-specific patch lengths, and text prototypes describing local trends.

What it assumes

Assumes basic time-series observations and transformer tokenization, while defining how grouped points become model tokens.

What you can do after

Apply patching before embedding to reduce token counts and retain local time-series information.

How it connects

Connects time-series forecasting with tokenization, embeddings, transformers, and text reprogramming.

liveBookmanning.com

Read Time Series Forecasting Using Foundation Models on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Build a Text-to-Image Generator (from Scratch)An image patch is a spatially localized portion of an image represented separately in a grid or latent representation.
• 

  liveBookmanning.com
  Deep Learning with R, Third EditionAn image patch is a localized rectangular portion of an image processed as an individual visual input unit.
• 

  liveBookmanning.com
  Deep Learning with Python, Third EditionA local subregion of an image that is extracted and treated as one sequence element in ViT processing.
• 

  liveBookmanning.com
  How Large Language Models WorkA localized rectangular portion of an image converted into a numerical representation for processing by a model.

Compare how 5 books explain patch →

Related concepts

• Vision Transformerused by · A Vision Transformer is a Transformer-based neural-network architecture that processes images as sequences of tokenized regions using attention.
• PyTorch Unfoldused by · PyTorch Unfold is a tensor operation that extracts sliding local blocks from an input tensor.
• embeddingproduces · A numerical vector representation of an item or input that encodes features or relationships useful for computational tasks.
• encoderproduced by · An encoder is a component that transforms input data into an encoded representation for subsequent processing, reconstruction, comparison, or transmission.
• VQGAN decoderproduced by · A VQGAN decoder is the image-reconstruction component that maps discrete image tokens to an output image.

This explainer was generated by Manning's knowledge graph from the text of Time Series Forecasting Using Foundation Models. The section links point to the parts of the book it draws on.