English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Time series modeling techniques based on deep learning have seen manyadvancements in recent years, especially in data-abundant settings and with thecentral aim of learning global models that can extract patterns across multipletime series. While the crucial importance of appropriate data pre-processingand scaling has often been noted in prior work, most studies focus on improvingmodel architectures. In this paper we empirically investigate the effect ofdata input and output transformations on the predictive performance of severalneural forecasting architectures. In particular, we investigate theeffectiveness of several forms of data binning, i.e. converting real-valuedtime series into categorical ones, when combined with feed-forward, recurrentneural networks, and convolution-based sequence models. In many non-forecastingapplications where these models have been very successful, the model inputs andoutputs are categorical (e.g. words from a fixed vocabulary in natural languageprocessing applications or quantized pixel color intensities in computervision). For forecasting applications, where the time series are typicallyreal-valued, various ad-hoc data transformations have been proposed, but havenot been systematically compared. To remedy this, we evaluate the forecastingaccuracy of instances of the aforementioned model classes when combined withdifferent types of data scaling and binning. We find that binning almost alwaysimproves performance (compared to using normalized real-valued inputs), butthat the particular type of binning chosen is of lesser importance.

The Effectiveness of Discretization in Forecasting: An Empirical Study on Neural Time Series Models