lightning-bolts VS Kedro

In the lightning-bolts repository, they implement the different RL algorithms, such as PPO and DQN, as different models. Would it make more sense to have the different algorithms be the Trainer instead? Inside each of the implementations, the model creates the same neural network with different training steps.

This is also how lightning bolts tend to be defined: for example, you have https://github.com/Lightning-AI/lightning-bolts/blob/master/pl_bolts/models/rl/advantage_actor_critic_model.py for A3C, which itself is only the loss and training wrapper around the normal modules for critic and actor (see https://github.com/Lightning-AI/lightning-bolts/blob/52e4c503c671f4866339c1537cf6ae506e7c5cf5/pl_bolts/models/rl/common/networks.py#L147=)

Also PyTorch Lighting gives example in this repo https://github.com/PyTorchLightning/pytorch-lightning-bolts

Interesting, thanks for sharing. I'll definitely take a look, although at this point I am so comfortable with Snakemake, it is a bit hard to imagine what would convince me to move to another tool. But I like the idea of composable pipelines: I am building a tool (too early to share) that would allow to lay Snakemake pipelines on top of each other using semi-automatic data annotations similar to how it is done in kedro (https://github.com/kedro-org/kedro).

# pyproject.toml [project] dependencies = [ "kedro @ git+https://github.com/kedro-org/kedro@3ea7231", "kedro-datasets[pandas.CSVDataSet,polars.CSVDataSet] @ git+https://github.com/kedro-org/kedro-plugins@3b42fae#subdirectory=kedro-datasets", ]

So there's 2 sides to pipeline management: the actual definition of the pipelines (in code) and how/when/where you run them. Some tools like prefect or airflow do both of them at once, but for the actual pipeline definition I'm a fan of https://kedro.org. You can then use most available orchestrators to run those pipelines on whatever schedule and architecture you want.

We have set up a milestone on GitHub so you can check in on our progress and contribute if you want to. To suggest features to us, report bugs, or just see what we're working on right now, visit the Kedro projects on GitHub.

you could use a project framework like kedro to force you to be more disciplined about how you structure your projects. I'd also recommend checking out this book: Edna Ridge - Guerrilla Analytics: A Practical Approach to Working with Data

Recently I started a position as Developer Advocate for Kedro, an opinionated data science framework, and one of the things we're doing is exploring what are the best open source tools we can use to create our documentation.

You can also check out Kedro, it’s like the Flask for data science projects and helps apply clean code principles to data science code.

An often overlooked framework used by NASA among others is Kedro https://github.com/kedro-org/kedro. Kedro is probably the simplest set of abstractions for building pipelines but it doesn't attempt to kill Airflow. It even has an Airflow plugin that allows it to be used as a DSL for building Airflow pipelines or plug into whichever production orchestration system is needed.