search

SakanaAI / AI-Scientist

The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery 🧑‍🔬
Apache License 2.0
7.91k stars 1.09k forks source link

run_experiment() still seems to be just executing the original experiment.py file #106

Open RemixaWorld opened 1 month ago

RemixaWorld commented 1 month ago

runexperiment() still seems to be just executing the original experiment.py file instead of 'run{run_num}.py' or something like that, and I don't see any code changes. Or where should I learn about aider's changes to the original code or parameter attempts?

https://github.com/SakanaAI/AI-Scientist/blob/2c41a92902cc40669059283a18d5923ff03fc8b2/ai_scientist/perform_experiments.py#L30

RemixaWorld commented 1 month ago

Or, I used pycharm to compare run_1.py/run_2.py/run_3.py/... It shows that they are all the same, which makes me wonder, if aider makes changes to experiment.py, where should I see the changed experiment.py explicitly? Maybe I made a mistake, please point it out for me. Anyway, thanks for contributions!

conglu1997 commented 1 month ago

It seems like Aider failed? Can you share your logs?

RemixaWorld commented 1 month ago

@conglu1997 I am sorry that I did not save the log. In my impression, aider only gave the experiment plan, but in fact, it did not perform any changes in the operation of experiment.py, that is to say, there was no search-replace record in the log. But when I ran it again today without changing anything, this time aider had made some change to experiment.py, which felt more like an erratic behavior of LLM. And, the model I used was deepseekv2, and maybe to this day, it has become deepseekv2.5.

conglu1997 commented 1 month ago

Hard to diagnose anything without any logs, but glad it works now.