Custom Interfaces with blocks

2022-12-07 · Source: Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Intermediate, medium

Summary

Prodigy's "blocks" feature enables the creation of custom annotation interfaces by combining existing UI components within a single recipe. This capability addresses the need to merge distinct annotation tasks, such as audio segmentation and text transcription, which are typically handled by separate Prodigy recipes like `audio_manual` and `audio_transcribed`. A custom recipe demonstrates how to define a list of dictionaries, each specifying a `view_id` for a pre-existing component (e.g., `audio_manual`, `text_input`), and then pass this configuration to the main recipe. The overall recipe's `view_id` must be set to "blocks" to render these combined components, allowing for a unified interface that captures multiple types of annotations, like speaker segments and corresponding text transcripts, in a single output JSON.

Key takeaway

For data scientists building custom annotation workflows, Prodigy's "blocks" feature offers a flexible way to combine disparate annotation tasks like audio segmentation and transcription into a single interface. This avoids switching between recipes, streamlining your labeling process and ensuring all relevant data is captured efficiently within one task. Explore the "view_id" and "blocks" configuration to tailor interfaces precisely to your project's needs.

Key insights

Prodigy's "blocks" feature allows combining existing UI components to build custom, multi-task annotation interfaces.

Principles

• Prodigy's "blocks" feature constructs interfaces from pre-existing components.
• Each component is defined by a "view_id" in a list of dictionaries.
• Custom recipes must set the main "view_id" to "blocks" and pass the block definitions.

Method

To create a custom interface, define a list of dictionaries, each with a "view_id" referencing a Prodigy component (e.g., "audio_manual", "text_input"), and pass this list as "blocks" in the recipe's config, setting the main "view_id" to "blocks".

In practice

• Combine "audio_manual" for segmentation with "text_input" for transcription.
• Add an "id" to text components to name the output field in annotations.

Topics

• Prodigy
• Custom Annotation Interfaces
• Audio Annotation
• Data Labeling
• Blocks Feature
• Recipe Development

Best for: AI Engineer, Machine Learning Engineer, Data Scientist

Related on AIssential

• Building Multimodal Corpora Using Microtask Pipelines and Local Annotators — Repurposing and enhancing existing commercial tools can efficiently build robust multimodal annotation infrastructure.
• PyCon Colombia Speaker Interview — Production-ready NLP tools and quality data annotation are crucial, with domain expertise being more vital than ML for its own sake.
• Advanced Personal Voice Activity Detection through Attention Score module with Conformer Block and FiLM Layers — Integrating Attention Score, Conformer Block, and FiLM Layers significantly enhances personal voice activity detection.
• Generative UI in Angular: Rendering AI-Chosen Components in Chat — Hashbrown enables Angular applications to integrate generative UI, allowing LLMs to render existing design system components.
• A multi-modal dataset for insect biodiversity with imagery and DNA at the trap and individual level — MassID45 integrates bulk and individual insect imagery with DNA data to train high-throughput taxonomic classifiers.
• Live Q&A With the Hosts: Measuring Progress Toward AGI - Cognitive Abilities Hackathon — Robust AGI evaluation requires targeted benchmarks for specific cognitive faculties, moving beyond general problem-solving.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai.