This repository contains end-to-end machine learning notebooks made by students in the 696DS course that showcase different functionalities with Microsoft Azure Machine Learning Services.
- Task: Classifying commericial blocks on telvision News Channels
- Methods: Comparing standard Scikit Learn models with Azure Automated Machine Learning
- Azure Functionalities: Running an Experiment, Automated Machine Learning, and Logging Metrics
- Results: Increased testing accuracy by 3.25%!
- Task: Image Classification on the CIFAR 10 dataset
- Methods: Fully connected neural network model
- Azure Functionalities: Hyperdrive run, Metric logging
- Results: Increased model classification accuracy by ~16%!
- Task: Classifying customers for customer retention problem.
- Methods: Model with highest accuracy- RandomForest
- Azure Functionalities: Model explanation
- Results: Understood the features responsible for someone's churn.
- Task: Translating sentences from German to English
- Methods: Using pre-trained BERT representations in Transformer Model
- Azure Functionalities: Register DataStore, AML Compute, Submitting and Cancelling Experiment Runs
- Results: 26.3
- Note: Implementation closely follows this tutorial
- Task: Predicting popularity of online news articles based on number of shares
- Methods: Classification
- Azure Functionalities: Auto Machine Learning (AML), AutoMLExplainer, Register Model
- Results: Iterated over 10 models using AML module. Achieved 67% accuracy over 5 different classes of labels (much higher than random!).