Manage data with
For data analysis, and especially machine learning, it is extremely valuable to be able to reproduce different versions of analyses that have been carried out with different data sets and parameters. However, in order to obtain reproducible analyses, both the data and the model (including the algorithms, parameters, etc.) must be versioned. Versioning data for reproducible analysis is a bigger problem than versioning models because of the size of the data. Tools like DVC help manage data by allowing users to transfer it to a remote data store using a Git like workflow. This simplifies the retrieval of certain versions of data in order to reproduce an analysis.
DVC was developed to be able to use ML models and data sets together and to manage them in a comprehensible manner. It works with different version managements, but does not need them. In contrast to DataLad/git-annex, for example, it is not limited to Git as version management, but can also be used together with Mercurial, see github.com/crobarcro/dvc/dvc/scm.py. It also uses its own system for storing files with support for SSH and HDFS, among others.
DataLad, on the other hand, focuses more on discovering and consuming datasets,
which are then easily managed with Git. DVC, on the other hand, stores each step
in the pipeline in a separate
.dvc file that can then be managed by Git.
.dvc files, however, allow practical tools for manipulating and
visualizing DAGs, see, for example, visualisation of DAGs.
External dependencies can also be specified with dvc remote.
Finally, external dependencies can also be specified with Pipenv.
You have to explicitly state the extras. This can be
ssh the command
looks like this:
$ pipenv install dvc[ssh]
Alternatively, DVC can also be installed via other package managers:
$ sudo wget https://dvc.org/deb/dvc.list -O /etc/apt/sources.list.d/dvc.list $ sudo apt update $ sudo apt install dvc
$ brew install iterative/homebrew-dvc/dvc
The following example was created with a current DVC version (1.0.0a9), which partly uses a different syntax than earlier versions. You can currently (8th June 2020) only install this with pip:
$ pipenv install dvc[all]==1.0.0a9