I have a Dockerfile in which I create a python environment using miniconda. This environment implicity installs a tensorflow version through specifying a rasa_core version 0.11.11. So I build an image from this Dockerfile. Now if I create a container locally with docker, I find the tensorflow version to be 1.10. However if I use the image in an deployment on kubernetes, the container crashes and the logs show that code from tensorflow version 1.11 is the cause. How is this possible?