Practical storage guide

Disk analysis for data scientists: datasets, environments, and checkpoints

Keep a reproducible record of datasets and experiments before trimming storage. A derived artifact can still be the only record of a costly or nonrepeatable run.

Where the space goes

Checkpoints and models

Several checkpoints may be needed for comparison or recovery, even if one looks best today.

Environments

Recreating an environment can depend on packages or credentials that are no longer available.

Private data inventories

Paths in reports can reveal customer names, research subjects, or internal project structure.

A practical cleanup workflow

STEP 1

Measure the actual source

Scan local datasets in the Mac app or use the CLI on a Linux machine. Use SSH when the data belongs to a remote host.

STEP 2

Save a report

Use nulldisk scan PATH --json for a record of the measured scope. Check for partial results before comparing totals.

STEP 3

Confirm reproducibility

Retain raw data, experiment metadata, locked dependencies, and any checkpoints needed to reproduce conclusions.

STEP 4

Review removals

Use explicit source-aware actions and a verified archive. Review report paths before sharing with collaborators or AI services.

Tools to compare

NullDisk offers native Mac maps, SSH/cloud sources, and development cleanup, plus a separate CLI. Match each tool’s platform and features to the job.

NullDisk features → · CLI and terminal workflows →

Common questions

Are old model checkpoints disposable?

Only after you decide they are unnecessary and can be recovered or reproduced. Age and size are review signals, not sufficient evidence.

Can I automate reporting?

NullDisk’s CLI supports JSON/NDJSON output, schemas, scan budgets, and exit statuses for incomplete scans. Keep reporting separate from authorized cleanup.