polakowo / vectorbt

Find your trading edge, using the fastest engine for backtesting, algorithmic trading, and research.
https://vectorbt.dev
Other
4.45k stars 625 forks source link

Import Errors #725

Open Anasdec opened 5 months ago

Anasdec commented 5 months ago

I'm having an import issue for past few days and still not able to solve it.

It's to do with importing _broadcast_shape from numpy.

Below is the error I get Untitled

psychedde1ic commented 5 months ago

I also have an issue with numpy as a result import issue.

ImportError: cannot import name '_broadcast_shape' from 'numpy.lib.stride_tricks'

is there any way to fix it? thanks!

Anasdec commented 5 months ago

@psychedde1ic have you managed to get the issue solved?

Diapolo10 commented 5 months ago

In case any other poor souls come across this issue, I did some digging and the issue stems from changes in Numpy 1.26.

It seems this commit removed _broadcast_shape and moved it under np.lib._stride_tricks_impl._broadcast_shape. All main Numpy versions newer than 1.25.2 have this change.

So the solution for users would be to use Numpy version 1.25.2 or older. The developers of VectorBT would need to add this new import (if they really want to rely on undocumented parts not in Numpy's public API) if they want to continue supporting newer Numpy versions. Personally I'd recommend refactoring the code to only rely on the public parts of the API to avoid similar problems in the future, as a better long-term solution.

Agent-AA commented 4 months ago

I also came across this issue. As @Diapolo10 said, the _broadcast_shape function still exists; it just got moved to a different file. Instead of reverting to an older Numpy version, I fixed the issue by commenting out the problematic import statement in the reshape_fns.py file and manually copied and pasted the _broadcast_shape function from _stride_tricks_impl.py into reshape_fns.py. A better solution from the devs would still be appreciated.

polakowo commented 4 months ago

Thanks. Fixed in 0.26.2

ccrockatt commented 1 day ago

I installed 0.26.2 and did not run into this issue. I had another issue due to numpy version incompatibility but that was fixed by removing numpy and running pip install vectorbt again and that installed a compatible version.