If you have a trained machine or deep learning model and want to deploy it on our platform, check out the Model Deployment Guides where you'll see tutorials for popular frameworks such as Scikit-learn (including a hands-on video), PyTorch, and Tensorflow.
Learn how to create and publish your algorithmsGet Started Now
If you have a trained machine or deep learning model and want to deploy it on our platform, check out the Deploy Model Guides where you'll see tutorials for popular frameworks such as Scikit-learn, PyTorch, and Tensorflow.
Best Practices and Deep Dives for Algorithm Development and Model Deployment
Follow these tips to create the best algorithm profile possible.
How to handle errors between algorithms.
Create, Update, Publish, and Delete Algorithms via the Algorithmia API
Using CI/CD to continuously deploy your ML models via Jenkins, GitHub Actions, or GitLab CI/CD
When you create an algorithm through the API, Algorithmia’s internal SCM provider is used by default unless an external SCM configuration is specified explicitly. To use an external SCM provider as the source code repository host for your algorithms, a cluster admin must first set up the configuration. At a high level, the workflow looks like this:
In the past, when shipping algorithms into production on Algorithmia, it’s been a challenge to ensure the integrity of model files. It’s also been difficult to update models, as it’s often the case that model file paths are hard-coded into the algorithm source code itself. Until now, there’s never been a standardized way to manage the serialized model files and other data files that are used by algorithms. Enter the model manifest system.
Reloading your ML model file when it changes
You can query Algorithmia’s API to retrieve build logs for a given algorithm. We offer the following two routes for retrieving build logs; you may decide that one is more convenient based on your workflow or use case. The
/builds endpoint can be used to retrieve information for any algorithm build, while the
/versions endpoint will only list algorithm builds that’ve been published.
You can query Algorithmia’s API to retrieve error-log stack traces for a given algorithm. The response will contain errors from the past 7 days, constrained to 200 algorithm executions. In order to support a variety of automated workflow configurations, we provide two separate ways to identify an algorithm on your cluster in order to request the logs.
Controlling where the source code for your algorithm resides
Table of Contents
Matrix of supported algorithm runtime environments
Getting your algorithm on Algorithmia is easy. Find out how with this tutorial.
Markdown syntax for Algorithm descriptions.