As the research outcome of data analysis or a machine learning experiment, moves towards the deployment, it is very essential that results are reproducible. This also requires exact versioning of the dependencies and libraries that have been used.
For e.g. numpy=1.20.0
A different version of even standard packages such as numpy or pandas on research environment can break the code or produce different results on production environment. As a best practice, go to your conda environment (preferably specific to specific projects) and export a requirements yml file or txt file. Make sure to push this in your git repo for your peers in order to maintain the environment consistency.