Finish deep SCM tutorial

[eb8680]

Paper: https://proceedings.neurips.cc/paper/2020/hash/0987b8b338d6c90bbedd8631bc499221-Abstract.html

Tasks:

• Finish models from paper using Pyro's normalizing flows or FlowTorch
• Add data (MNIST?)
• Add inference Reproduce results in deep SCM tutorial #70
• Add evaluation Reproduce results in deep SCM tutorial #70
• Deep SCM Example Text #114 Add explanatory text
• Incorporate transform fixes from Pyro once they land Add conditional inverse and compose TransformModules pyro-ppl/pyro#3185
• Update and re-generate results once Support ambiguous conditioning on TransformedDistributions #108 lands on master
• Deep SCM Notebook: Download dataset from web source within notebook #179
• Update package dependencies to include any new packages used in the notebook

[eb8680]

@zenna do you want to help @archanarw take this one on?

[zenna]

Yes. I can help. I anticipate most of the questions will be Pyro-specific. @archanarw built https://github.com/archanarw/NormalizingFlows.jl and has a lot of experience with Omega. @archanarw let's go through the paper tomorrow

[SamWitty]

@eb8680, could you please update the task list above with the current status? Thanks!

[eb8680]

The notebook in #70 has complete results and visualizations, it just needs text. Since #70 can't be merged into master until #97 lands, you can put up PRs for the notebook against #70.