using-django-docker-and-scikit-learn-to-bootstrap-your-machine-learning-project/
summary
Claims that Django can spin up Docker containers that build and version models, using Docker images as reproducible snapshots containing code data and a pickled scikit‑learn model. Workflow uses Jupyter inside a container, scikit‑learn pipelines and Joblib pickles, and a Django endpoint that triggers Docker image builds and returns image URLs; a simple admin surfaces metrics like false positives/negatives. Example data: 5,671 requests 994 labeled successful, and repositories with the Dockerfile and Django wrapper exist and run as described.
AI-written summary · about
chapters (19)
- 0:00Start
- 0:29Recap
- 5:39What Is Machine Learning
- 6:25Machine Learning
- 8:04Performance Metric
- 15:21Data Science Team
- 21:38Where Is the Handoff between Data Science and Production
- 23:04Using Scikit-Learn
- 26:54Reusability
- 28:33Purple Heart Emoji
- 29:42Data Governance
- 31:37Why Use Docker
- 32:38Docker File
- 33:02Build the Docker Image
- 34:22Mountable Data Directories
- 36:00What I Could Do Is Use in Python We'Ve Got a Docker Module Which Allows Us To Create an Image so if I Were To Stand Up a Docker Endpoint in a in a Django Api Maybe the Url Looks Something like this Crate Image Name of Model Go Ahead Create Image in that That Will Kind Of Go Ahead and Go through the Procedure of Saying Hey I'M GonNa Read the Docker File I'M GonNa Go Ahead and Build this Image and I'M GonNa Go Ahead and Stuff It Somewhere and What I Want To Return Back Is Basically like a Url of Where of Where that Docker Image Lives
- 38:11So the Big Thing That I Did Not Talk about To Keep in Mind Is It's Not Just that We Want To Automate Kind of the Building of these Things but We Want To Understand How Models Will Do over Time Right that's the Big Thing Here That I Think Is Kind of the Elephant in the Room Is We Want To Be Able To Store Analytics As Well So Currently What My Team's Working on Is We'Re Using the Jingo Admin To Start Bringing In some of the Analytics That We Get with Scikit-Learn
- 38:45Training Set We Now Have Consistency It's like Okay Model Variant One Here's the Results of False Positive False Negative Model Variant Number Two Here's the Results Such that When We Start Looking at Maybe What's the Best Model We Can Start Surfacing those in the Django Admin View some Other Things To Kind Of Think about-Is There some Other Things To Think about Too Is We Haven't Gone the Path of Talking about Kubernetes in this Talk but Kubernetes Is a Great Way To Stand Up an Image and Stand It behind an Endpoint So Again Taking that Containers Technology to the Next Step and Allowing People Then Who Are on Who Are the Application Developers To Go Ahead Start Hitting an Endpoint Kubernetes Can Also Allow Us To Get to Work and Start Doing that
- 39:49I Have Also Used Falcon As Well I Think It Really Depends on What Your Team Needs but I Think Django Is a Really Good Place To Start cuz It's Got Really Dry Bus Support and Integration with a Lot of these Tools We'Ve Talked about So if You'Re Curious about Kind Of like What Next There's a Lot of Places We Can Go Next I Really Enjoy Rob Stories Talk Bridging Python to Jvm because that's the Other Thing Maybe We Prototype Remodel and Python but We Want To Make Use of the Jvm Is There Something We Can Do that Well Yeah We Can Definitely Do this with Our Flow We Just Might Need To Transform that Pickled Model into a Different into a Different Format How Can We Do that
related talks
45mDjango for AI: Deploying Machine Learning Models with Django with Will Vincent
54mRipped Machine Learning with Distributed Computing
35mGetting Machine Learning Models Ready For Production Using Python
23mAn Intro to Docker for Djangonauts
44mA Django Project Template with CI/CD Pipelines and k8s Orchestration
24mDocker for Machine Learning!
35mMachine Learning Pipeline using Luigi and Scikit Learn
22mProductive Machine Learning and Deep Learning Projects