Endless "python not initialised" warning when entering training mode

[samrma]

Endless "python not initialised" warning when entering training mode

[Muginson]

I am also having this exact issue, no idea what to do.

[mooch443]

Hey,
sorry to hear that you are experiencing these issues. May I ask when you installed this version? Since this command just installs the "current" version at all times. The loop is endless because it does not seem to terminate the program and instead tries again.

In general the error seems to indicate that the numpy version installed is too new for the code to run. Can you provide a conda list command output for your environment? You could try to install a 1.19 version of numpy (either using pip or preferably conda if that is possible). In order to do that you should simply run conda install numpy=1.19 from within the environment (make sure you're not in the base environment).

Potential problems with this might be... other software's versions incompatible. But in order to judge that, I'd first have to see what the environment looks like.

@samrma sorry for the late response. I had another response planned but it did not pan out, and then I was swamped :(
-Tristan

[samrma]

Thanks for your comment, Tristan - I appreciate your help with this. I assumed it was a compatibility issue but couldn't understand why - my experience with anaconda is limited (I don't use it for anything else) but I've only ever used the installation code provided (i.e., conda create -n tracking -c trexing trex), which presumably installs the correctly compatible packages. I attempted to install numpy 1.19 in my previous environment as suggested but got the same loop. Attached is the conda list for a fresh environment (named fishtracker) - I hope this helps. Thanks again!

[mooch443]

Curious. You got the same loop, even though the numpy installation was successful? Thats weird since numpy with versions below 1.20 should not even produce a warning here. Are you sure the installation succeeded?

I just tried a fresh install on Windows using:

conda create -n late_response_again --override-channels -c trexing -c defaults trex=1.1.9 numpy=1.23 tensorflow-gpu

and that seems to work through the entire visual identification pipeline. Notice I added --override-channels (just to be sure you dont have anything funny going on with channels on your PC), and added a numpy=1.23 (which will install "any" 1.23.x version of numpy). In my case it got 1.23.5. I also added tensorflow-gpu since it didn't install it automatically.

Edit: oops, nevermind. It installed an old CPU version of TRex instead. Updated the above command. In any case - the problem seems to be that you get 1.26.x on a fresh install, which is incompatible it seems.