In this repository, you'll find sample projects of integrating machine learning models with Arthur, with tips on how to get the most out of the Arthur Python SDK. Each example shows you how to prepare data, register models with Arthur, send inferences to the platform, and use Arthur features to analyze model performance.
Data scientists and machine learning practitioners use Arthur to get insights about their models' performance and centralize their organization's use of AI in one unified system for tracking and analysis. You can find more information about the Arthur SDK and API on our documentation site.
This sandbox contains many example projects of onboarding machine learning models to Arthur. Below, we describe which example makes sense for which user, and at the end provide a table to look up the features demonstrated in each example project.
For a quickstart, check out the Credit Card Default example project for a beginner-friendly example.
It walks through onboarding a tabular binary classification model to the Arthur platform and showcases the
full suite of Arthur features available to free plan users.
The MEPS Healthcare Utilization example project is another binary classification task, similar to the credit card default example project. The Boston Housing example project, a regression task that also demonstrates preparing the model with PySpark.
With an enterprise account, you can use Arthur for Computer Vision (CV) and Natural Language Processing (NLP) models.
Check out our Mars Rover example project, which receives images from an actual rover on Mars via the NASA API, and applies a computer vision model to perform image classification.
Three other examples show how to use Arthur with images and computer vision models. The Cancer Detection example project and Satellite Images example project both demonstrate image classification tasks, similar to the Mars rover example project. The Object Detection example project demonstrates using Arthur to monitor models that objects within images.
In addition, the Medical Transcripts example project demonstrates an NLP classification task.
To install the Arthur Python SDK, you can use the pip package manager, and run pip install arthurai at your command
line.
For example, Arthur offers monitoring services for models used in Computer Vision (CV) and Natural Language Processing (NLP) applications. See the table below for a reference of examples in this repository.
| Example | Input Type | Output Type | Inference Type | Data Storage | Model Storage | Arthur Features | Frameworks |
|---|---|---|---|---|---|---|---|
| Credit Card Default | Tabular | Binary | Batch & stream | Local | Local | Performance, Drift, Visualization, Alerts, Bias, Explainability, Anomaly Detection, Hotspots, Bias Mitigation | Scikit-Learn |
| MEPS Healthcare | Tabular | Binary | Stream | Local | Local | Explainability | Scikit-Learn, NLTK |
| Boston Housing | Tabular | Regression | Batch | Local | Local | Explainability | Spark ML |
| Cancer Detection | Image | Binary | Stream | Arthur S3 Bucket | Local | Explainability | PyTorch |
| Mars Rover | Image | Object Detection | Stream | Nasa API and Arthur S3 Bucket | SageMaker | Explainability | Scikit-Learn, OpenCV, PyTorch |
| Object Detection | Image | Object Detection | Stream | Arthur S3 Bucket | Local | Explainability | OpenCV, TensorFlow |
| Satellite Images | Image | Binary | Batch | Arthur S3 Bucket | Local | Explainability | PyTorch |
| Medical Transcripts | NLP | Multiclass | Stream | Local | Local | Explainability | Scikit-Learn, NLTK |
| Headline Summarization | NLP | Generative Text | Stream | Huggingface SDK | Huggingface SDK | Coming Soon! | Huggingface |
| Comparing LLMs | NLP | Generative Text | Stream | Huggingface SDK | Cohere & OpenAI API Endpoint | Coming Soon! | Huggingface, Cohere, OpenAI |
| OpenAI GPT3 | NLP | Generative Text | Stream | Huggingface SDK | OpenAI API Endpoint | Coming Soon! | Huggingface, OpenAI |
| OpenAI ChatGPT | NLP | Generative Text | Stream | Huggingface SDK | OpenAI API Endpoint | Coming Soon! | Huggingface, OpenAI |