ChatGPT is a hot topic and I wanted to play around with it a bit.
Most code here has been created using ChatGPT.
_prompts.md are the prompts used for the respective scripts.
Some scripts had to be slightly fixed, but not a lot. All visible in the git history.
The purpose of the scripts was to get a small toolbox to understand deep learning's approximation capabilities for functions with one input and one output.
In trainings you can find the resulting plots, usually showing training, testing and predicted data with overlaid scatter plots.
Conclusion:
I think chatGPT is a great tool for prototyping scripts. You need to precise in your description but don't have to bother about syntax. This makes the progress faster. Sometimes, the generated scripts contain mistakes or unexpected default values. Took me about 2 hours to get the first plots with prediction results. That's easily the time I usually have for the first training script or so.