With reference to my thread (https://julius.ai/s/0b5f8102-df57-49e9-8d95-894d864eaa5c), when requesting the AI to compile the python code from the start to the end of the analysis, it starts to compile the code starting only after the ETL/ELT steps (using the cleaned data) and ignores the compile for all the previous ELT/ETL steps that it has gone through.
Please advice how to get it to compile all codes from the start of the thread to the end, I have tried several ways including saving the thread as a workflow/notebook, but the only way was to manually back track.
It depends on how much code you’re trying to compile, but Julius should be able to handle most cases. Double check that you have extended memory on – have seen it work on compiling the code that was ran in a thread / notebook into a json file.
Eg. Can you compile all of the code that was ran in this conversation into a json file?
Hi Tyler, Thanks for responding. I was working continously on the same thread within the same day, for about 3 hours. I prompted the thread with “compile all python code so far”, but it missed out all ETL steps.
If I go back to the thread now, and I can still see all the codes all the way from the start, and I tried prompting with “Can you compile all of the code that was ran in this conversation into a json file?”. - this did not work. It still picked up python codes only after certain steps as per before.
The other advice you gave was having the extended memory on. But since all the codes are still visible right now, i do not understand how extended memory will help, other than keeping past tables/datasets?
Interesting - so it’s still missing out on the ETL steps even though you can view them.
Potentially could be an issue with Jupyter hub where your Julius code / conversations is executing. Even though you can see code in your thread / notebook, Julius might not actually have it in the environment (for whatever reason). That was one reason why extended memory could’ve helped, it allows Julius to look back in prior conversations better than without it.
One other thing I’d try is downloading your conversation as a Jupyter Notebook (something that was recently added). I haven’t tried with a very long conversation, but in theory you can download the entire conversation, and select the the code specifically.
In any case I’ll forward this thread to the product team – they’ll investigate to see if there’s any other reason why it’s having issues compiling the code.
I am able to download the entire python code using the option “download a Jupyter Notebook”. Thank you.
I have yet to try the extended memory to see if that could have helped, but since I did not take a break from using the thread, I was assuming that everything in the conversation is going to be intact? Am I wrong to assume this?
Awesome!!! Really glad to here - def will work on getting that added to our documentation!
It really does depend (and in this case does sound like a bug if you really didn’t step away.) Julius should’ve been able to see all the way back - but there could be any number of reasons Julius might lose some context. That’s thing that extended memory can help fix (in addition other benefits)
