Best AutoML systems based on tabular data?

Question

Analytics

Given: A
car of tabular data in the amount of approximately 5 GB in one MySQL database table.
This is a kind of log of user actions.
Task:
To try to make some prediction of the user's further actions.
Again, research! Test a few theories, try a few guesses.
Unfortunately, all programmers of the project work on php. Some people have Python-knowledge and very, very poor.
Tell me,
is there something simple like Google AutoML Tables but self-hosted? With some kind of outside access, maybe an API?
There are no problems with training - we can allocate 4x1080ti for research. I'm sure more than enough.

Answer 1

Somehow very messy. And it seems to me that the approach is not from the other side.
Let me explain.
1. ML is not PHP, Python or C++. This is primarily knowledge and experience in applying methods - as you write - "foresight" (including, of course). Therefore, the experience of your programmers is the last thing you should care about if you are interested in "I emphasize again - research"
2. Well, suppose you find something like self-hosted AutoML Tables. Now we read "AutoML Tables enables your entire team of data scientists, analysts, and developers to automatically build and deploy state-of-the-art machine learning models on structured data at massively increased speed and scale."That is, first of all - DS specialists and analysts! Do you have them? If you have, then ask them what tools they want to work with. If not, then .... miracles do not happen.
3. That's when you will select DS-specialists when they think about how to analyze your data, perform Feature engineering, select (choose) at least a class of methods that make sense to apply, conduct a pilot study of the data, approximately evaluate their prospects - and it may well ( and most likely it will be so) it turns out that AutoML capabilities are not enough for your task - then you will have to talk about video card farms and the experience of specific developers
. not a meat grinder - a piece of meat at the entrance, minced meat at the exit.
4. And so, there are enough AutoML systems on the market. Well, offhand - H2O AutoML, Auto-WEKA, TransmogrifAI, Firefly, etc. Here is the latest (in time) review
https://www.datasciencecentral.com/profiles/blogs/...
and good links for further reading.
PS And, by the way, 5 GB of logs is a very modest amount, especially for any type of AutoML.