conda environment.yml

[mateuszwyszynski]

We need a proper requirements for conda installation. Currently the installation through the requirements.txt file does not work and the "from scratch" installation mixes installation though pip and conda. This is ok, but:

• maybe we can install everything using conda - this is most likely the most recommended scenario, because we use one package manager
• if not, we should at least do the installation using conda and pip in the officially recommended order, i.e. first install as much as possible through conda and only then start installations using pip

[mateuszwyszynski]

Moreover, there are some additional installs which are necessary (they pop up when trying to run different scripts) and are not covered during the installation process:

• [ ] tensorboard
• [ ] ipdb
• [ ] sklearn

[mateuszwyszynski]

Ok, as far as I know smplx cannot be installed through conda, so pip has to be used. Nevertheless we should create a working environment.yml file. The current version works on my machine, but it does not work on VM instances on GCP. On the date of this comment, there is new a section in INSTALL.md which goes through the steps needed to install everything on GCP (and in general from scratch on, in theory, any machine). But this is still messy.

Most likely all of the comments above will have to be revised when I'm done with cleaning up the code. Hopefully we will be able to throw out some packages. Or at least separate things used for presenting (e.g. like the current jupyter install I believe) from the things needed for the model itself.