Data-specific model

Q: How deeply does Time Series Forecasting Using Foundation Models cover data-specific model?

Demonstrates data preparation, model fitting, held-out prediction, plotting, and MAE and sMAPE evaluation.

As explained in Time Series Forecasting Using Foundation Models

A forecasting model designed and trained for a particular dataset, domain, or use case rather than broad general-purpose use.

A data-specific model is trained for one dataset, domain, or use case, specializing in its particular patterns rather than serving many situations. section 1.0

In forecasting, fit the model directly to the target series and its frequency. This can outperform a pretrained model when the data contains granular or domain-specific patterns, but requires a separate model for each situation. section 1.0 section 2.8

Evaluate a data-specific model by comparing it with a pretrained alternative:

Use a held-out test horizon and compare forecasts against actual values. section 2.8
Measure errors with metrics such as MAE and sMAPE rather than relying only on plots. section 2.8
Prefer it when the pretrained model misses the dataset’s frequency or domain-specific behavior. section 1.3.2 section 2.8

Daily temperature forecasting

A pretrained model trained only on monthly data produced daily forecasts with MAE 2.59 and sMAPE 8.69%. An N-BEATS model trained on the daily dataset achieved MAE 1.34 and sMAPE 4.87%. section 2.8

Listing 2.16 Training a data-specific model on the daily data

models = [NBEATS(input_size=2*horizon, h=horizon, max_steps=500)]

nf = NeuralForecast(models=models, freq='D')
nf.fit(df=d_input_df)
d_trained_preds = nf.predict()

Shows the minimal NeuralForecast setup for fitting an N-BEATS model specifically to daily data. section 2.8

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

foundational modelcontrasts with · A base or foundation model is a pretrained model that provides learned parameters or capabilities as a starting point for reuse or adaptation to downstream tasks.
N-BEATSincludes · N-BEATS is a neural forecasting model organized as stacks of blocks for modeling time series without requiring explicit series decomposition or input transformations.
symmetric mean absolute percentage errormeasured by · Symmetric mean absolute percentage error is a forecasting metric that measures absolute prediction error relative to the magnitudes of predicted and actual values.
zero-shot forecastingalternative to · Zero-shot forecasting is forecasting with a model on a target time series without task-specific training or fine-tuning on that series.

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.