This blog covers some basics, experiments and related math in the field of Machine Learning [ML]. It is a personal blog and not an ordered book. Contents comes with numerical experiments I had some fun with.
I write in general about experiments which one can perform on a medium equipped Linux PC. Meaning: This blog will mainly cover conventional experiments which can e.g. be done with Scikit-Learn and Neural Networks with a rather limited number of layers. Still, I think that one can learn quite a lot of interesting things from such limited experiments.
Besides the fun factor: One can prepare oneself via studying some basics for bigger and more professional tasks.
For the time being this post is not yet about GPT and other advanced transformer based neural networks. The reason is simply that I need a new graphics card to perform related experiments. I will order one soon.
Who is this blog for?
I expect this blog to be interesting for people who have already started with private ML projects – but are no experts, yet. There is a variety of standard experiments one typically starts with. You will sooner or later find such experiments with variations here in this blog. But I also intend to cover some experiments which you may not find in introductory text books. So, the posts will cover topics both for beginners and advanced users of Python, Numpy, Scikit-Learn Keras and Tensorflow. I will try to point out what level of knowledge may be required to understand a post or a post series.
You are invited to ask questions, write comment and exchange experiences. However, I expect that you open an account on this blog and let me check your comments before publishing them.
Equipment to do your own experiments
If you want to do similar projects as discussed here you should be prepared to have some 32 GB RAM and a Nvidia card with at least 4 to 8 GB of VRAM. My personal programming environment are Jupyter Lab (for Python) and Eclipse with PyDev. I strongly advice you not to work with Jupyter, only. Instead you should systematically gather and reorder your work with neural networks systematically within classes and reusable methods. And you should collect your classes in suitable Python modules. An Eclipse/PyDev environment in my opinion is much more suitable for such tasks than Juypter.
I do all my ML experiments on Linux systems. Please, do not expect me to answer questions regarding PyDev and Jupyter installations on Windows.
What may distinguish this post from others is that I sometimes will write about mathematical aspects I stumble across during my experiments and which I find interesting. I will try to confine posts within a separate main category.
Most of the mathematical subjects I have so far looked into deal with linear algebra (matrix operations), some features of statistical multivariate normal distributions, ellipsoids and ellipses.
Further topics will follow.
The role of my linux-blog
Some people may know me from my linux-blog hosted at anracom.com. In the linux-blog I wrote about Linux- and LAMP-related topics the first years (up to 2014). During the last 10 years, however, the linux-blog has become a container for all kind of IT-topics.
Among other things it got a growing section for Machine Learning. As some readers of the linux-blog have recently complained about an overload of only partially Linux-related topics I have opened this new blog. I intend to transfer selected ML-related posts from the linux-blog to this new blog.