Data scarcity

Q: How deeply does Time Series Forecasting Using Foundation Models cover data scarcity?

Explains scarcity’s causes and mitigation through dataset aggregation, frequency balancing, and varied pretraining windows.

As explained in Time Series Forecasting Using Foundation Models

Data scarcity is a condition in which too little suitable data is available for training or analysis.

Data scarcity is the lack of sufficient suitable data for training or analysis. In foundation forecasting, open time-series data is especially scarce: natural-language datasets are far larger, and combining series across industries and sources is difficult. This limits which teams can build large forecasting models and can constrain model quality. section 2.9

Large-scale forecasting work addresses scarcity by:

Compiling LOTSA, an open archive containing more than 27 billion observations across energy, transportation, climate, cloud operations, web, sales, nature, economics, and health care. section 6.1.2
Subsampling overrepresented datasets so models do not overspecialize in dominant frequencies, especially hourly data. section 6.1.2
Varying input lengths and forecast horizons during pretraining to support series with different temporal structures. section 6.1.2

Table 6.2 Breakdown of LOTSA by frequency

Frequency	Number of datasets	Number of observations
Yearly	4	0.87 million
Quarterly	5	2.31 million
Monthly	10	11.04 million
Weekly	7	18.48 million
Daily	21	709.02 million
Hourly	31	19.88 billion
Every minute	25	7.01 billion
Every second	2	14.79 million

Use the frequency breakdown to inspect how observations are distributed across LOTSA and why sampling is needed for balanced generalization. section 6.1.2

How Time Series Forecasting Using Foundation Models covers it

Cover of Time Series Forecasting Using Foundation Models

liveBook manning.com

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

Related concepts

classifieraddressed by · A classifier is a predictive computational model that assigns inputs to one or more discrete classes or categories.
sparsitycontrasts with · Sparsity is the property of data, a representation, or a structure in which relatively few possible entries, relationships, or components are present, nonzero, or active.

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.