Is there any plans to release a kedro light? Or a ...
# questions
j
Is there any plans to release a kedro light? Or a version without the project management. I really like the package but I have a hard time putting this into production with my conda environments for work because kedro has some heavy requirements like specific
git
versions which I'm assuming is for the project management features kedro supplies...or
cookiecutter
which needs to be downgraded from the latest anaconda release
y
Hi John, thanks for the question! Let me see if I can try help you here. If I summarise your question, you're asking us if we could release a version of Kedro with fewer dependencies? At the moment, our list of dependencies is indicated here. Are
cookiecutter
and
gitpython
the reasons you are struggling? What's causing the issue?
j
Yes you are correct. Interesting observation, I'm attempting to install this release of kedro on conda-forge.
conda-forge/noarch:: kedro-0.18.3-pyhd8ed1ab_0 None
When doing so, conda wants to do the following
(see attachment below)
Deleted
Deleted
The conda-forge version wants to install perl and a new git, which I'm confused about with respect to your requirements.txt
Note I already have gitpython 3.1.27 installed
my_env.txt
kedro-dry-run.txt
To make things easier to understand. An explicit listing of my current conda env and my dry run of installing kedro from conda-forge
To restate my question, I see that kedro is planning on making datasets it's own package, is there a plan to break the DAG aspect out? I find kedros DAG framework to be the most alluring aspect. The project management aspect is nice too but I do not have a need for it and it seems to facilitate the need for a lot of extra dependencies.
When I say DAG framework I mean the
pipeline
,
node
, and
datacatalogue
objects. At work we made a DAG framework and are debating on doing a full replacement with kedro's. You all have made a much simpler api that seems more fleshed out than ours.
y
Ah! There is a PR out to address the `conda-forge` issue by @Ahdra Merali and @Merel. And then to your query around using Kedro without the project template, this is possible. Kedro is a framework but it's also a library too. The visual in our architecture diagram best shows this, and you'll see that the
pipeline
,
configloader
,
runner
and
i/o
(data catalog) are library components. We actually do show how to use Kedro this way in our "Hello Kedro" example; and our API docs for each component show how to leverage them too. You mentioned that you don't want the project template, why would that be? Do you have your own?