Quantile quantization

As explained in Machine Learning System Design

A transformation that converts continuous values into discrete buckets defined by value quantiles.

Quantile quantization transforms continuous features into categorical features by grouping values into discrete buckets. Use it when a model benefits from discrete inputs rather than raw measurements. section 11.1.2

DeepETA feature pipeline

Uber’s DeepETA network transforms continuous, categorical, and geospatial data into discrete tokens, then into learnable embeddings for a transformer-based ETA model. section 11.1.2

Figure 11.1 An overview of the DeepETA model pipeline: example of combining base feature engineering and a deep learning model (source: https://mng.bz/0MoN)

Use the pipeline figure to trace how feature engineering turns mixed tabular inputs into representations consumed by the deep learning model. section 11.1.2

How Machine Learning System Design covers it

How deep it goes

Defines value-based bucketing and shows an embedding pipeline, but does not detail bucket-count selection or implementation steps.

Examples

Uses Uber’s DeepETA network to illustrate transforming mixed tabular data into discrete tokens and learnable embeddings.

What it assumes

Assumes familiarity with continuous features and feature transformations while introducing their conversion into categorical buckets.

How it connects

Connects feature engineering with categorical features, embeddings, transformer architecture, and model-dependent transformations.

liveBookmanning.com

Read Machine Learning System Design on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  D3.js in Action, Third EditionA quantile scale maps continuous input values to discrete output values according to thresholds that divide the input distribution into groups of similar sizes.

Compare how 2 books explain quantile quantization →

Related concepts

• d3.scaleQuantile()implemented by · d3.scaleQuantile() is a D3-scale constructor that creates a quantile scale from a discrete input domain and discrete output range.
• inputproduces · An input is a value, information, or signal supplied to a system, computation, or interface for processing or to influence its behavior.
• empirical cumulative distribution functionuses · An empirical cumulative distribution function gives each value the proportion of observed samples less than or equal to that value.
• quantizationkind of · Quantization maps numerical values to a finite set of discrete representations, often using reduced precision to lower model storage or computation requirements.
• upper quartileuses · The upper quartile, or third quartile, is the value marking approximately the 75th percentile of an ordered dataset.
• discrete variableuses · A discrete variable is a variable whose possible values belong to a finite or countable set of distinct alternatives rather than a continuous range.

This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.