There could be a few things. The notebook path goes through a symlink a few levels up the file system (before hitting the repo root, though). I'm also in an anaconda env with a mix of conda and pip packages. The notebook kernel is also not the default kernel, but that's a common use case.

If it helps to know, I'm running on Linux with jupyterlab, and the trains-server is hosted on the same machine (using the defaults from the docker-compose.yml ). I'm on trains version 0.16.3 .