lazydata: scalable data dependencies
lazydata is a minimalist library for including data dependencies into Python projects.
Problem: Keeping all data files in git (e.g. via git-lfs) results in a bloated repository copy that takes ages to pull. Keeping code and data out of sync is a disaster waiting to happen.
lazydata only stores references to data files in git, and syncs data files on-demand when they are needed.
Why: The semantics of code and data are different - code needs to be versioned to merge it, and data just needs to be kept in sync.
lazydata achieves exactly this in a minimal way.
- Keeps your git repository clean with just code, while enabling seamless access to any number of linked data files
- Data consistency assured using file hashes and automatic versioning
- Choose your own remote storage backend: AWS S3 or (coming soon:) directory over SSH
lazydata is primarily designed for machine learning and data science projects.
See this medium post for more.
In this section we'll show how to use
lazydata on an example project.
Install with pip (requires Python 3.5+):
$ pip install lazydata
Add to your project
lazydata, run in project root:
$ lazydata init
This will initialise
lazydata.yml which will hold the list of files managed by lazydata.
Tracking a file
To start tracking a file use
track("<path_to_file>") in your code:
from lazydata import track # store the file when loading import pandas as pd df = pd.read_csv(track("data/my_big_table.csv")) print("Data shape:" + df.shape)
Running the script the first time will start tracking the file:
$ python my_script.py ## lazydata: Tracking a new file data/my_big_table.csv ## Data shape: (10000,100)
The file is now tracked and has been backed-up in your local lazydata cache in
~/.lazydata and added to lazydata.yml:
files: - path: data/my_big_table.csv hash: 2C94697198875B6E... usage: my_script.py
If you re-run the script without modifying the data file, lazydata will just quickly check that the data file hasn't changed and won't do anything else.
If you modify the data file and re-run the script, this will add another entry to the yml file with the new hash of the data file, i.e. data files are automatically versioned. If you don't want to keep past versions, simply remove them from the yml.
And you are done! This data file is now tracked and linked to your local repository.
Sharing your tracked files
To access your tracked files from multiple machines add a remote storage backend where they can be uploaded. To use S3 as a remote storage backend run:
$ lazydata add-remote s3://mybucket/lazydata
This will configure the S3 backend and also add it to
lazydata.yml for future reference.
You can now git commit and push your
lazydata.yml files as you normally would.
To copy the stored data files to S3 use:
$ lazydata push
When your collaborator pulls the latest version of the git repository, they will get the script and the
lazydata.yml file as usual.
Data files will be downloaded when your collaborator runs
my_script.py and the
track("my_big_table.csv") is executed:
$ python my_script.py ## lazydata: Downloading stored file my_big_table.csv ... ## Data shape: (10000,100)
To get the data files without running the code, you can also use the command line utility:
# download just this file $ lazydata pull my_big_table.csv # download everything used in this script $ lazydata pull my_script.py # download everything stored in the data/ directory and subdirs $ lazydata pull data/ # download the latest version of all data files $ lazydata pull
lazydata.yml is tracked by git you can safely make and switch git branches.
Data dependency scenarios
You can achieve multiple data dependency scenarios by putting
lazydata.track() into different parts of the code:
- Jupyter notebook data dependencies by using tracking in notebooks
- Data pipeline output tracking by tracking saved files
- Class-level data dependencies by tracking files in
- Module-level data dependencies by tracking files in
- Package-level data dependencies by tracking files in
- Examine stored file provenance and properties
- Faceting multiple files into portable datasets
- Storing data coming from databases and APIs
- More remote storage options
Stay in touch
This is an early stable beta release. To find out about new releases subscribe to our new releases mailing list.
The library is licenced under Apache-2 licence. All contributions are welcome!