<aside> 🔗
Please use the links below to quickly navigate the Notion Page.
<aside>
</aside>
<aside>
</aside>
<aside>
</aside>
<aside>
</aside>
<aside>
</aside>
<aside>
Link to Test it (api.insight4data.com)
</aside>
</aside>
<aside> 📝
Purpose
The MLapp project aims to bring machine learning and modeling capabilities to people of all skill levels who want a step-by-step, visual way to create, update, deploy, and run batch inference on models. Each screen has one feature — such as variable selection, scaling, model selection, and outlier removal — and is designed in a way where new features and machine learning model support can be added in sequence.
The MLapp also currently offers support for a few types of graphs to compare model accuracies and residuals to provide quick and easy visual comparison graphs. Please refer to the Pharmaceutical Formulations project for more graphs that will be introduced in the near future.
Method
A full list of the software and databases used are in the Software Details section of the Technical Outline. The app uses Django for the frontend and backend, AWS S3 for static and media files generated, and ZenML and MLflow to handle logging, registering, and deploying models. All 3 servers are run on docker containers inside of the EC2 instance, with nginx used to route traffic between each container. The The interface guides you through importing you data as a CSV to creating a model. Based on the data in the CSV, the app will also list out some basic suggestions for models that can be used as a guide. Once the model is created and a “good” model is found, it can be deployed and used to make single live predictions as a persistence HTTPS endpoint, or to make batch inferences that can be stored.
Outlook
The most versatile part of the project is the ability to add more features. The ability to create and compare machine learning models prior to deployment in a visual interface will bring more comprehensive solutions to all industries. Currently it works for Scikit models, and the premise can be easily expanded to start supporting Tensorflow models as well.
</aside>
<aside> 📽️
</aside>
<aside> 📝
</aside>