Launch HN: DAGWorks – ML platform for data science teams

• hamilton

  20 1,312 9.8 Jupyter Notebook

  Hamilton helps data scientists and engineers define testable, modular, self-documenting dataflows, that encode lineage and metadata. Runs and scales everywhere python does.

  

Hey HN! We’re Stefan and Elijah, co-founders of DAGWorks (https:///www.dagworks.io). We’re on a mission to eliminate the insane inefficiency of building and maintaining ML pipelines in production.

DAGWorks is based on Hamilton, an open-source project that we created and recently forked (https://github.com/dagworks-inc/hamilton). Hamilton is a set of high-level conventions for Python functions that can be automatically converted into working ETL pipelines. To that, we're adding a closed-source offering that goes a step further, plugging these functions into a wide array of production ML stacks.

ML pipelines consist of computational steps (code + data) that produce a working statistical model that a business can use. A typical pipeline might be (1) pull raw data (Extract), (2) transform that data into inputs for the model (Transform), (3) define a statistical model (Transform), (4) use that statistical model to predict on another data set (Transform) and (5) push that data for downstream use (Load). Instead of “pipeline” you might hear people call this “workflow”, “ETL” (Extract-Transform-Load), and so on.

Maintaining these things in production is insanely inefficient because you need both data scientists and software engineers to do it. Data scientists know the models and data, but most can't write the code needed to get things working in production infrastructure—for example, a lot of mid-size companies out there use Snowflake to store data, Pandas/Spark to transform it, and something like databrick's MLFlow to handle model serving. Engineers can handle the latter, but mostly aren't experts in the ML stuff. It's a classic impedance mismatch, with all the horror stories you'd expect—e.g. when data scientists make a change, engineers (or data scientists who aren’t engineers) have to manually propagate the change in production. We've talked to teams who are spending as much as 50% of their time doing this. That's not just expensive, it's gruntwork—those engineers should be working on something else! Basically, maintaining ML pipelines over time sucks for most teams.

One way out is to hire people who combine both skills, i.e. data scientists who can also write production code. But these are rare and expensive, and in our experience they usually are only expert at one side of the equation and not as good at the other.

The other way is to build your own platform to automatically integrate models + data into your production stack. That way the data scientists can maintain their own work without needing to hand things off to engineers. However, most companies can't afford to make this investment, and even for the ones that can, such in-house layers tend to end up in spaghetti code and tech debt hell, because they're not the company's core product.

Elijah and I have been building data and ML tooling for the last 7 years, most recently at Stitch Fix, where we built a ML platform that served over 100 data scientists from various modeling disciplines (some of our blog posts, like [1], hit the front page of HN - thanks!). We saw first hand the issues teams encountered with ML pipelines.

Most companies running ML in production need a ratio of 1:1 or 1:2 data scientists to engineers. At bigger companies like Stitch Fix, the ratio is more like 1:10—way more efficient—because they can afford to build the kind of platform described above. With DAGWorks, we want to bring the power of an intuitive ML Pipeline platform to all data science teams, so a ratio of 1:1 is no longer required. A junior data scientist should be able to easily and safely write production code without deep knowledge of underlying infrastructure.

We decided to build our startup around Hamilton, in large part due to the reception that it got here [2] - thanks HN! We came up with Hamilton while we were at Stitch Fix (note: if you start an open-source project at an employer, we recommend forking it right away when you start a company. We only just did that and left behind ~900 stars...). We are betting on it being our abstraction layer to enable our vision of how to go about building and maintaining ML pipelines, given what we learned at Stitch Fix. We believe a solution has to have an open source component to be successful (we invite you to check out the code). In terms of why the name DAGWorks? We named the company after Directed Acyclic Graphs because we think the DAG representation, which Hamilton also provides, is key.

A quick primer on Hamilton. With Hamilton we use a new paradigm in Python (well not quite “new” as pytest fixtures use this approach) for defining model pipelines. Users write declarative functions instead of writing procedural code. For example, rather than writing the following pandas code:

  df['col_c'] = df['col_a'] + df['col_b']

• docs.getdbt.com

  2 106 10.0 JavaScript

  The code behind docs.getdbt.com

  

allowing users to write imperative code (e.g. using loops) that dynamically generates DAGs are never a good idea. I say this as someone who personally used to pester framework PMs for this exact feature before. While things like task groups (formerly subDAGs) [2] appear initially to be right answer, I always ended up regretting them. They're a scheduling/orchestration solution to a data transformation problem

Can y'all speak to how Hamilton views the data and control plane, and how it's design philosophy encourages users to use the right tool for the job?

p.s. thanks for humoring my pedantry and merging this! [3]

[1]: https://github.com/dbt-labs/docs.getdbt.com/pull/2390

• WorkOS

  workos.com sponsored

  The modern identity platform for B2B SaaS. The APIs are flexible and easy-to-use, supporting authentication, user identity, and complex enterprise features like SSO and SCIM provisioning.

  

• quokka

  23 1,081 8.3 Python

  Making data lake work for time series (by marsupialtail)

would love to collaborate on an integration with pyquokka (https://github.com/marsupialtail/quokka) once I put out a stable release end of this month :-)

• awesome-pipeline

  10 5,904 5.7

  A curated list of awesome pipeline toolkits inspired by Awesome Sysadmin

As a long-time fan of DAG-oriented tools, congrats on the launch. Maybe you can get added here https://github.com/pditommaso/awesome-pipeline now or in the future...

This is a problem space I've worked in and been thinking about for a very, very long time. I've extensively used Airflow (bad), DBT (good-ish), Luigi (good), drake (abandoned), tested many more, and written two of my own.

It's important to remember that DAG tools exist to solve two primary problems, that arise from one underlying cause. Those problems are 1) getting parallelism and execution ordering automatically (i.e. declaratively) based on the structure of dependencies, and 2) being able to resume a partially-failed run. The underlying cause is: data processing jobs take significant wall-clock time (minutes, hours, even days), so we want to use resources efficiently, and avoid re-computing things.

Any DAG tool that doesn't solve these problems is unlikely to be useful. From your docs, I don't see anything on either of those topics, so not off to a strong start. Perhaps you have that functionality but haven't documented it yet? I can imagine the parallelism piece being there but just not stated, but the "resumption from partial failure" piece needs to be spelled out. Anyway, something to consider.

A couple more things...

It looks like you've gone the route of expressing dependencies only "locally". That is, when I define a computation, I indicate what it depends on there, right next to the definition. DBT and Luigi work this way also. Airflow, by contrast, defines dependencies centrally, as you add task instances to a DAG object. There is no right answer here, only tradeoffs. One thing to be aware of is that when using the "local" style, as a project grows big (glances at 380-model DBT project...), understanding its execution flow at a high level becomes a struggle, and is often only solvable through visualization tools. I see you have Graphviz output which is great. I recommend investing heavily in visualization tooling (DBT's graph browser, for example).

I don't see any mention of development workflow. As a few examples, DBT has rich model selection features that let you run one model, all its ancestors, all its descendants, all models with a tag, etc etc. Luigi lets you invoke any task as a terminal task, using a handy auto-generated CLI. Airflow lets you... run a single task, and that's it. This makes a BIG DIFFERENCE. Developers -- be they scientists or engineers -- will need to run arbitrary subgraphs while they fiddle with stuff, and the easier you make that, the more they will love your tool.

Another thing I notice is that it seems like your model is oriented around flowing data through the program, as arguments / return values (similar to Prefect, and of course Spark). This is fine as far as it goes, but consider that much of what we deal with in data is 1) far too big for this to work and/or 2) processed elsewhere e.g. a SQL query. You should think about, and document, how you handle dependencies that exist in the World State rather than in memory. This intersects with how you model and keep track of task state. Airflow keeps task state in a database. DBT keeps task state in memory. Luigi track task state through Targets which typically live in the World State. Again there's no right or wrong here only tradeoffs, but leaning on durable records of task state directly facilitates "resumption from partial failure" as mentioned above.

Best of luck.

Suggest a related project