LIONoso for "Learning from data"
by Caltech professor Yaser Abu-Mostafa
LIONlab believes in top-quality education for everybody.
An exciting initiative in this direction is the
Learning From Data (Introductory Machine Learning) course by Yaser S. Abu-Mostafa,
Professor at the California Institute of Technology.
Among LIONlab team there are people who were Yaser's students years ago.
As a small concrete step to "give back to the community," a
free LIONoso license is available to all course
participants, for the entire course duration (tools details here).
In addition, the book:
THE LION WAY: Machine Learning plus Intelligent Optimization
is available to all course participants in digital form at a symbolic fee
(and free for participants in Kindle Unlimited).
The LION way gives a complementary view with respect to the book by Yaser
and colleagues. In essence, they focus more on the theoretical underpinnings of
"learning from data", we focus more on the practical aspects of machine learning
and on the interplay between machine learning and intelligent optimization
Yaser's course is not about giving you pre-digested baby food, but about strengthening your teeth.
The best way to learn will be to dirty your hands to develop some software, make mistakes, debug, etc.
To this end we suggest you use programming languages of your choice (including "fast" scripting languages like
Python) or whatever language you prefer
for the core part of the homework and then connect modules to LIONoso for visualizations, plots, more advanced
and open-ended explorations.
After the homework is due, we encourage exchanging the additional explorations with the community.
Every week we plan to place in this web page specific software exercises related to the course.
You may want to bookmark it (http://lionoso.org/learningfromdata/).
LIONoso makes it easy to test the principles and methods presented
in the course on real examples. Most operations happen by drag and drop with immediate visual output.
In addition to basic exercises, it encourages open-ended and far-reaching personal
experimentation, ranging from machine learning to optimization.
The software includes tools to:
- Import data from files and from most existing databases, either local or distributed in
different servers (and reachable via secure web connection).
- Use exploratory data analysis tools to rapidly navigate in your data, visualize them and
derive the relevant summaries (including bar charts, bubble charts, pie charts, histograms,
radar charts, line plots, parallel coordinates, and advanced capabilities like sweeps in time or across parameters).
- Use model-building tools, including supervised learning methods like neural networks,
unsupervised learning methods like clustering (top-down and bottom-up), self-organizing maps.
- Access a suite of solvers (optimizers) matched to the business characteristics.
- Manipulating data tables, for preparing training and validation files, merging tables together,
filtering data, etc.
- Design experiments (DOE) for creating input data to be used for testing your business (or your system).
LIONoso runs on Windows and Mac OS X platforms.
Linux support is experimental, and may not work in all systems: Wine
has been suggested as an alternative way to use LIONoso on Linux systems.
To install the software for the course follow the steps in the standard
LIONoso download page.
Have fun with the course! If you encounter difficulties, read the accompanying
read the manual contained in the software
come back to this page, or,
if none of the above works, contact us
(specify "learningfromdata" in the Title box)
so that your question will immediately reach our colleagues in charge. Everybody involved is
a volunteer so
please be kind :)