added get-started-experiments generation code

[iesahin]

This PR contains the source code and generation script for https://github.com/iterative/get-started-experiments

• This is split of [WIP] A new GS project based on MNIST #35
• Related with get-started-X: Write generate scripts for each project #34

It includes the general generate.bash script that prepares the build directory, calls the generate-experiments.bash and writes a push script that pushes the tags to the repository.

• merge generate.bash and generate-experiments.bash in a single file and move the directory one level up
• push the resulting models/fashion-mnist/model.h5 to remote
• check and update the remote per repos: don't use dataset-registry to store models, intermediate artifacts, etc #48
• prepared and preprocessed data should not be downloaded from remote (mnist examples: simplify data layout, params, etc #54)
• merge prepare and preprocess stages to reduce the intermediate file size and use the cache (Comment in mnist examples: simplify data layout, params, etc #54)
• remove fashion-mnist intermediate directory from data and models to reduce the final depth
• rename the directory to get-started-experiments
• update the Readme file
• simplify initial .gitignore
• move venv instructions in a <detail> block
• modify according to this review

[casperdcl]

Initial review (mostly just *.md) for now.

• Since you use vim, can you use a plugin which strips trailing spaces (line ending) as well as end-of-file (single-newline)?
• Also don't use [links][like this] unless you re-use them. Otherwise it's just a liability (broken/missing links) - spotted at least one I think.

[jorgeorpinel]

May want @dberenbaum's review instead of mine in this one 🙂

Just one minor Q: is "generate.bash" Bash-specific or is it a general Shell script (.sh extension) ?

[iesahin]

* Since you use `vim`, can you use a plugin which strips trailing spaces (line ending) as well as end-of-file (single-newline)?

Actually I used VS Code with Remote editing in this PR. Vim has these config, I'll take care of this.

[iesahin]

* Also don't use `[links][like this]` unless you re-use them. Otherwise it's just a liability (broken/missing links) - spotted at least one I think.

I'm using these because of readability. I read these documents usually as text. They fit into the paragraph and don't distract the flow. Broken links are a different matter, though, I don't put the links while writing and forget to put them sometimes. They are easy to spot though.

[casperdcl]

It looks like this PR has already been deployed to https://github.com/iterative/get-started-experiments before being reviewed & merged 🤔

Anyway for all these generation-type stuff we should also have a note at the bottom of generated READMEs linking back to this source repo.

[iesahin]

I generated and pushed the repos because I think it's easier to see the end product in review. I see the generation code as an internal stuff while the resulting repos as the product, but I may add the links to this repo if it feels better.

[casperdcl]

I generated and pushed the repos because I think it's easier to see the end product in review.

best to push to a temp repo rather than an in-production repo (otherwise defeats at least one of the objectives of review)

I see the generation code as an internal stuff while the resulting repos as the product, but I may add the links to this repo if it feels better.

exactly. Advanced users (and us too, for our own sanity) care about internal stuff being linked (a one-line contributing section stating "please open PRs in the source repo if you find a bug," etc.)

[shcheklein]

Nice @iesahin ! It's way simpler now.

params.py now contains a flat list of 4 parameters: instead of hierarchical model.cnn.conv_units, we now have conv_units.

I would keep one nested level (to show that it's possible)

Also a few other comments:

• can we use images and labels instead of .gz?
• simplify the code also to some reasonable extent

[shcheklein]

But overall, I would try at this point to write some documentation first and see what can we get there - then do other iterations on this.

[iesahin]

I think this is done. I'm about to merge if no more changes requested. @shcheklein @jorgeorpinel @casperdcl

[shcheklein]

preferable use the same style across all *.md

[shcheklein]

some minor comments here and there, looks good otherwise to me.

[iesahin]

@shcheklein About the root .gitignore file I probably deleted the file mistakenly because I never had any intention to work on files outside of the get-started-experiments directory. I've just updated the README recently. I've reinstated the global .gitignore.