Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Don’t Democratize Data Science (builtin.com)
32 points by loumal on May 8, 2020 | hide | past | favorite | 5 comments


So I do not think that this is a black/white situation.

I do agree, that any auto ml tool, cannot know what is the correct data sources for a specific use case, nor does it know the business case value of a specific model.

Auto ML should not replace humans, but work with them. I.e. it is a tool.

This tool can do the following:

1. Auto visualization.

2 Auto feature engineering.

3. Auto backtesting.

4. Auto training (with cost optimization).

5. Hyperparameter optimization.

6. Auto model packaging.

7. Autoload testing and auto security testing.

7. Auto deploy / scale / SLA.

8. Auto monitor.

The goal of AutoML is to save time and reduce risk. I am not sure why you want to do all of the above manually.


I'm all about AutoML but the best data scientists do what the author says: They try to understand the problem in detail first. I think fewer and fewer do this well.


Absolutely, auto ml cannot know your problem, as a word processor cannot know what book you want to write.

The goal of AutoML is to compress the time/money from idea/problem def to running models, it is not to replace the data scientist.


it’s a counterintuitive argument but I completely agree


Can you imagine trying to democratize phd level mathematics? This boot camp shit has gotten insane.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: