DVC initialisieren und einen Datensatz verfolgen

DVC is a Python package, so a single pip command gets you started. After this you have the dvc command available in your terminal.

DVC works alongside Git, not instead of it. So you run it inside an existing Git repository where your code already lives.

dvc init adds a hidden .dvc folder holding config and the local cache. Commit it once so the whole team shares the same setup.

Point DVC at the file or folder you want versioned with dvc add. DVC moves the real data into its cache and leaves a small placeholder behind.

dvc add creates a tiny .dvc file next to your data. It stores the file's hash and size, acting as the address Git will track.

You commit the small .dvc file to Git, never the raw dataset. Git now records which version of the data this commit expects.

So you never commit the big file by accident, dvc add auto-adds the dataset path to .gitignore. Git simply ignores the heavy bytes.

DVC names cached files by their content hash. Identical data is stored once, so duplicate datasets never waste disk space. 💾

Run dvc status any time to see whether your working data matches what the .dvc files expect. It is your quick health check.

When the dataset changes, just run dvc add again. DVC recomputes the hash and updates the .dvc file, which you commit as a new version.

dvc add works on whole directories, not just single files. Point it at a folder and DVC versions everything inside as one tracked unit.

After running dvc add on a dataset, what do you commit to Git?

You ran dvc init, then dvc add to track a dataset. DVC caches the real data and writes a tiny .dvc pointer that Git commits, keeping code and data linked.