dagshub-project-demo/notes.txt

- No Private-Public Key Encryption
- To use an access token for git cloning, use this address:
    https://<access token>@dagshub.com/path/to/your/repo.git
- Note: only https-level authorization, no SSH-level cloning
- Uses DagsHub storage for storing models, data
- Some issue with DVC requires me to downgrade to dvc==2.3.0
    -- https://discord.com/channels/698874030052212737/705302775784800278/890961327462416435
- Some issue installing the particular version of Numpy
- key thing: Track Data Files with DVC, Track code with Git
- Can render tabular data within the DagsHub UI
- Tracking Experiments:
    -- DagsHub Logger save model metrics from experiments as .csv files, and
        saves parameters as .yaml files
    -- DagsHub Logger looks to be agnostic to the list of hyperparameters
        established for your model
    -- Model experiment is tied to a commit
- Testing a new hypothesis
    -- Creating a new benchmark model to compare is typically represented as a
        new branch
- Question:
    -- Does DVC support Model and Data Storage Management on S3 Buckets?
        --- Yes it does!
        --- https://dagshub.com/docs/integration_guide/set_up_remote_storage_for_data_and_models/
    -- What are the storage limits of DagsHub Storage?
        --- Up to 10GB of DAGsHub Storage for free model
        --- 1 TB for team model
        --- Unlimited for Enterprise quote
    -- What are these added blue columns to the experiment set?
- Potential Issues to raise on DagsHub repo:
    -- issue with numpy dependency in the tutorial
- Pricing Model:
    -- $0 for all public repos
    -- $0 for all private repos with up to 2 additional collaborators
    -- $49 for team subscription to get unlimited collaborators, 1 TB of
        DAGsHub storage
    -- Enterprise custom quote for anoything more
- Key Takeaways:
    -- Use Git to version your code, DVC to version your data
    -- Need to first git commit data.dvc, before committing your data via DVC
    -- An experiment is tagged to a commit
    -- We assume your best model sits in the master branch, and when you want
        to try a new experiment, you branch off of master for an experiment
    -- For DVC on DagsHub storage, Data does not look to be versioned by the
        code that created it
- How are models saved and versioned?
- Why DagsHub: https://dagshub.com/docs/faq/
    - DagsHub is a web platform for data version control and collaboration for
        data scientists and machine learning engineers.
    - Git is not good at versioning large files
    - git-lfs is an extension to git that can be used to version large files,
        but they don't version the data pipeline.
    - DagsHub is basically a UI on top of the Git-DVC source control flow
    - DagsHub represents each node as a file, with important details and a
        direct link to the file itself