awslabs/gluonts

Using dynamic categorical features in NN models

Open

#392 opened on Oct 15, 2019

View on GitHub
 (6 comments) (2 reactions) (0 assignees)Python (3,888 stars) (753 forks)batch import
enhancementgood first issuehelp wanted

Description

GluonTS allows to work with dynamic categorical features, but there is no way to use them in most of the NN models (e.g. DeepAR, Transformer, etc.)

They can be incorporated by applying embeddings and concatenating them with other dynamic features. Does it make sense to modify DeepAR and other models with this change, or is it better to stick with the current implementation and to write new model implementations that will use dynamic categorical features?

Contributor guide

Tech stack
pythonpytorch
Domain
machine learningdata
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
GluonTS basicsNeural network fundamentalsEmbedding layers
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40
Research direction
Investigate how dynamic categorical features are currently handled in GluonTS (e.g., in DeepAR and Transformer models). Propose adding an embedding layer for categorical features and concatenating them with existing dynamic features. Discuss with maintainers whether to modify existing models or create new ones. Review the model code in src/gluonts/model/ for relevant files.