Someone really needs to look at the research notebooks. There is very odd behavior happening and has been for a week now.
- Notebooks "mysteriously" can't reference variables or functions periodically, sometimes even in the same code block. This shows up as "blah blah is undefined" when you're staring right at it.
- Similarly, notebooks that run fine and have been used many times will suddenly "not work" for odd reasons. This takes numerous forms but the one I've seen most recently was that "estimate /= num" is improper syntax. A few reloads of the kernel does not fix it. Changing the line, eg. to "estimate = estimate / num" fixes it and magically I can change it right back to "estimate /= num" and it suddenly works again.
- Memory measurement seems off. I can run a notebook that uses ~60% of the memory, rerun a simple plotting cell and it will reload as being out of memory. I don't know what is causing this but it's incredibly irritating to lose hours or days of results to something like this.
- When the second item happens, I will sometimes get inconsistent numerical results. I have started double calculating certain things (once with numpy, once with straight python) so I can detect when this is happening and reload the kernel.
If people are experiencing any of the above behavior, be very careful about the last item and put some sanity checks in. Until this is resolved, I would be highly suspicious of any results in the research notebooks unless you have triple checked everything and can validate the results across multiple kernel restarts.
My suspicion is these are all related. Is anyone else experiencing anything like this?