A smörgåsbord of AutoML Frameworks (AMLFs). If you ask me what an AMLF does, I wouldn’t have an exact answer for you. Well, it depends on each AMLF; each one has a different flavour and taste. So before I munch an AMLF, I’d ask 2 questions:
Q1: What is this AMLF automating? Here’s a tentative list of ML tasks that could be automated:
Hyperameter Optimisation (HPO)
Neural Architecture Search (NAS)
Meta-Learning for NAS and HPO
Feature engineering, selection
Model evaluation, selection via model competitions
ML pipelines with genetic programming
ML workflows like data pre-processing
Cost-optimisation of AutoML model compute
ML specialised tasks, e.g. regression for xyz…
Explainability of AutoML-generated models
Augmentation or elimination of ML code
Q2: How can I compare AMLFs? Obviously with so many potential features to automate and so many AMLFs out there, comparing AMLFs is extremely tricky. Luckily we can use AMLB, a benchmark that compares 8 main AMLFs. It also defines best practices and provides recommendations to avoid common mistakes when comparing AMLFs.
Extensible benchmark. The great thing is that AMLB is extensible, so you can add more datasets, constraints, and AMLFs. I sort of miss some notable exceptions from the original benchmark comparison. Here is a list of AMLFs that you could add:
Interested in the latests on AMLFs? Then I suggest you read the papers from the AutoML-Conf 2022, 1st ever conference on this.
10 Link-o-Troned
A Pythonista *Experience*
beCause of Dennis & Bjarne
Scripting aRt
Love from Julia
(Paren(th)ethical)
ScalaTOR
data v-i-s-i-o-n-s
Distributed de-Entangler
Forschung!
Algorithmic Potpourri
Robots & Cyborgs like <you>
Deep & Other Learning Bits
startups -> radar
ML Datasets & Stuff
Postscript, etc
Tips? Suggestions? Feedback? email Carlos
Curated by @ds_ldn in the middle of the night.