Skills for ML Engineering

Posted on

1. Getting ready

  1. Do a lot of leetcode problems to understand data structures and algorithms
  2. Get exposure to software design patterns (e.g. event driven etc.)
  3. Get solid CS fundamentals by reading text books (e.g. Operating Systems, Compilers, Distributed and cloud computing)
  4. Get solid ML engineering fundamentals https://course.fast.ai/ or https://www.coursera.org/professional-certificates/tensorflow-in-practice or https://www.udacity.com/course/deep-learning-pytorch—ud188.

2. Learn from Google

Google is an industry leader in ML and surrounding best practices. What they have to say about the topic is useful.

  1. Do the crash course - https://developers.google.com/machine-learning/crash-course
  2. Read the People+AI guidebook - https://developers.google.com/machine-learning/guides
  3. Read Rules of ML - https://developers.google.com/machine-learning/guides/rules-of-ml
  4. Read the other guides - https://developers.google.com/machine-learning/guides
  5. Try the TensorFlow quickstart in Google Colab - https://www.tensorflow.org/tutorials/quickstart/beginner
  6. Read the TFX Guide and maybe try it out - https://www.tensorflow.org/tfx/guide

3. Write a pipeline

  1. If you have access to an AWS account, run a complete SageMaker pipeline https://github.com/aws/amazon-sagemaker-examples/tree/main/sagemaker-pipelines/nlp/amazon_comprehend_sagemaker_pipeline and go around your AWS console seeing and inspecting all the resources it created
  2. If you have access to a Google Cloud account, run a Vertex AI pipeline https://cloud.google.com/vertex-ai/docs/pipelines/build-pipeline
  3. If you have Azure, then https://docs.microsoft.com/en-us/azure/machine-learning/how-to-create-machine-learning-pipelines
  4. If you don’t have access to any account already, poke around the internet if you can figure out if any of them offer things in free tier (it’s also a skill to see if a vendor provides something in free tier and use that I guess ;)). If not, run a Kubeflow pipeline https://www.kubeflow.org/docs/components/pipelines/introduction/, or build an Airflow pipeline https://www.google.com/search?q=pytorch+airflow

4. Research & explore the web

Just surf the internet, maybe start with https://github.com/visenger/awesome-mlops and check out what others are up to, especially companies like Uber, AirBnB, Lyft, Shopify, Netflix, etc.

FBLearner flow, particularly, a post from Facebook all the way back from 2016 is still as awesome as ever and is a brilliant design pattern introduction for doing ML at a big company https://engineering.fb.com/2016/05/09/core-data/introducing-fblearner-flow-facebook-s-ai-backbone/.

5. Andrew Ng

Do the Andrew Ng course at https://www.coursera.org/specializations/machine-learning-engineering-for-production-mlops. Just do it.

6. Talks and conferences

Develop a method to search the Internet and find people speaking about the topic. A lot of people speak about ML these days in different forums, events and conferences.

  1. Stanford MLSys seminars https://www.youtube.com/playlist?list=PLSrTvUm384I9PV10koj_cqit9OfbJXEkq
  2. DeepLearning AI channel https://www.youtube.com/watch?v=Ta14KpeZJok
  3. AWS re:Invent 2021 keynote https://www.youtube.com/watch?v=ue9aumC7AAk
  4. A podcast I found on InfoQ’s channel https://www.youtube.com/watch?v=aPbQMx7zuOM
  5. The awesome-mlops index again https://github.com/visenger/awesome-mlops#talks-about-mlops
  6. My own talk at AWS re:Invent from 2019 https://youtu.be/fgLlEnhFZQA?t=1200

Discuss this post on HN