Deployment

What’s a Dark Launch

• Continuously deliver with dark launches and feature toggles
• What Is A Dark Launch?
• CRE life lessons: What is a dark launch, and what does it do for me?
• Dark Launching: A Way to Test New Features Before Going Live
• The Only Guide to Dark Launching You’ll Ever Need

Considerations for Production

• 7 Considerations Before Pushing Machine Learning Models to Production
• Shipping to Production

Over-the-wall Data Science

• Over-the-Wall Data Science and How to Avoid Its Pitfalls

MLOps Courses

• MLOps Zoomcamp
• MLOps Zoomcamp Summary

Docker build with -f option cannot find Dockerfile

Well, as the error message is suggesting, your Dockerfile is not within the context, which is the current directory (.) in your case.

The docker file, specified with -f, must always be within the context directory specified as an argument.

So, normally this should work fine:

docker build -f /path/to/context/dir/Dockerfile /path/to/context/dir

And this too:

cd /some/dir
docker build -f /some/dir/customDir/Custom-Dockerfile-name .

While, this would fail:

docker build -f /path/to/diff/dir/Dockerfile /path/to/context/dir

From the Docs:

docker build [OPTIONS] PATH | URL | -

The docker build command builds Docker images from a Dockerfile and a “context”. A build’s context is the set of files located in the specified PATH or URL

And:

By default the docker build command will look for a Dockerfile at the root of the build context. The -f, –file, option lets you specify the path to an alternative file to use instead. This is useful in cases where the same set of files are used for multiple builds. The path must be to a file within the build context. If a relative path is specified then it is interpreted as relative to the root of the context.

Source: https://stackoverflow.com/questions/32235987/docker-build-with-f-option-cannot-find-dockerfile

Model Retraining

The Ultimate Guide to Model Retraining

A guide on when to retrain your Machine Learning model

Monitoring

• Monitoring Machine Learning Systems

SageMaker Workflows

• SageMaker Workflows

Bayesian vs Frequentist - Bayesian is Computationally Expensive because of integration

• Technically Wrong: When Bayesian and Frequentist methods differ

Machine Learning Operations (MLOps): Overview, Definition, and Architecture

• Machine Learning Operations (MLOps): Overview, Definition, and Architecture

Does Model Performance Even Matter?

Source: Does Model Performance Even Matter? - YouTube - Ritvikmath

TL;DR
Q: What makes a model “good”?/How to choose which model to use?
A: Depends on what it’s going to be used for.

A non-exhaustive list of metrics we need to consider depending on the model’s purpose:

• Performance (accuracy, precision, etc.)
• Training volume
• Training time
• Inference time
• Interpretability
• Model storage (size in MB/GB)

Consider trade-offs for your situation and environment

• K-NNs take up a lot of disk space because they hold the entire dataset. So, if storage is limited, go for a different model.
• If model HA has higher accuracy than LA, but Longer Inference time than LA and if our most important consideration is how quickly we can show the result rather than how accurate it has to be, go for model LA and vice-versa.
• Neural Networks take a long time to train because they have a lot of parameters that need to be updated. They’re also hard to explain and interpret.
• Whereas, Linear Regression is fast to train and also easy to explain and interpret.
• However, a Neural Network may perform better (eg. higher accuracy) than Linear Regression.

Operationalizing Machine Learning: An interview study

https://arxiv.org/pdf/2209.09125.pdf