Please enable JavaScript.
Coggle requires JavaScript to display documents.
Automated Machine Learning (Eight criteria for AutoML excellence (Accuracy…
Automated Machine Learning
Machine learning life cycle
Not a linear process
First step
Defining product objectives
Specify business problem
Aquire subject matter expertise
Define unit of analysis
Define prediction target
Identify success criteria
Foresee risks
Decide whether to continue
Any machine learning that automates repetitive tasks required for effective machine learning
Automating complex processes
Beyond cognitive ability of humans
AirBnB
Want to build customer lifetime value models for guests and hosts
Make decisons about individual hosts as well as aggregated markets such as any city
Increase division efficiency
Identified 4 repetitive tasks where AutoML positively impacted productivity
Exploratory data analysis
Feature engineering
Algorithm selection and hyper parameter tuning
Model diagnosis
What automated learning is not
It is not automatic ML
still several decisions that must be
made by the analyst and a requisite skillset for evaluating the results of applying
machine learning to any data set
Generally
speaking, AutoML is good news for data scientists because it frees them from
manually testing out all the latest algorithms
Available tools and platforms
Context based tools
Implemented within another system
For a specific purpose
General platforms
General purpose machine learning
Open source
tools tend to be developed by and for
computer and data scientists and generally require knowledge of programming
languages
Commercial
Provided by commercial vendor for a price
Eight criteria for AutoML excellence
Accuracy
Most important criteria
Stems from system selecting which features to use and creating new ones automatically
Compares and selects variety of relevant models and tunes models automatically
Validation process
Resource compatibility
Should work with existing business systems
Easily integrated
Productivity
Process transparency
Improving the knowledge of AutoML
Ease of use
Generalizable across contexts
Should work for all data types, sizes and different times
Understanding and learning
Should improve analysts understanding
Should visualize the interactions between feature and target
Recommended actions
Mostly for context specific AutoML
