The machine learning process: From ideation to deployment with Azure Machine Learning
- 2. Useful Resources: @frlazzeri ✓ AzureML GitHub: aka.ms/AzureMLrepo ✓ Algorithm Cheat Sheet: aka.ms/AlgorithmCheatSheet ✓ Deep Learning VS Machine Learning: aka.ms/DeepLearningVSMachineLearning ✓ Automated Machine Learning Documentation: aka.ms/AutomatedMLDocs ✓ Model Interpretability with Azure ML Service: aka.ms/AzureMLModelInterpretability ✓ Azure Machine Learning Service: aka.ms/AzureMLservice ✓ Azure Machine Learning Designer: aka.ms/AzureMLdesigner ✓ Get started with Azure ML: aka.ms/GetStartedAzureML ✓ Azure Notebooks: aka.ms/AzureNB
- 5. idea reality data experiment model deploy @frlazzeri
- 6. Azure Data Lake Azure Data Factory Azure DataBricks Azure SQL Azure Machine Learning GitHub TensorFlow, PyTorch, sklearn Azure Compute – CPU/GPU/FPGA Azure DevOps GitHub Azure Kubernetes Service Azure IoT Edge Azure Monitor Data Engineer Data Scientist ML Engineer There are many jobs & tools involved in production ML IT / Ops Data Analyst Business Owner & many more… @frlazzeri
- 7. Train Model Train Evaluate Model Registry Release Model Validate DeployPackage Profile Approve Data Lake Data Catalog Data Engineer Featurize Data Scientist release collect ML Engineer register Prepare Data E2E Machine Learnin process @frlazzeri
- 8. The Machine Learning Process @frlazzeri
- 9. @frlazzeri
- 10. Store 1 Balancing capital investment constraints or objectives and service-level goals over a large assortment of stock-keeping units (SKUs) while taking demand and supply volatility into account. Store 2
- 11. Business Understanding @frlazzeri ✓ Ask the right questions ✓ Define Performance Metrics ✓ Understand what to do with your data
- 12. Understand what to do with your data aka.ms/AlgorithmCheatSheet
- 13. Business Understanding Ask the right questions What are the forecasted sales quantities per item per store for the next 4 weeks? Using forecasting models such as determining reorder points and economic order quantities can help ensure optimal inventory control. Define Performance Metrics Evaluation metric: MAPE Understand what to do with your data Regression – Time Series Forecasting approach Data-driven stock, inventory, ordering Omni-channel shopping experience with machine learning Predict inventory positions and distribution Fraud detection Market basket analysis Demand plans Forecasts Sales history Trends Local events/weather patterns Inventory optimization @frlazzeri
- 14. Data Acquisition & Understanding @frlazzeri ✓ Data Architecture ✓ Data preprocessing ✓ Feature Engineering
- 15. Stores Information about the 45 stores, indicating the type and size of store Features Contains additional data related to the store, department, and regional activity for the given dates. • Store - the store number • Date - the week • Temperature - average temperature in the region • Fuel_Price - cost of fuel in the region • MarkDown1-5 - anonymized data related to promotional markdowns. • RDPI - Real Disposable Personal Income • Unemployment - the unemployment rate Sales Historical weekly sales data, which covers 3 years: • Store - the store number • Dept - the department number • Date - the week • Weekly_Sales - sales for the given department in the given store POS data
- 16. Step 1 Build and train Step 2 Package and deploy Step 3 Monitor POS data Power BI Inventory ordering system Data Lake Storage Azure Synapse Analytics Azure Machine Learning Forecasting @frlazzeri
- 17. Find Features and Process Data
- 18. Find and Validate Data Starting Dataset
- 19. Add RDPI Index to data Resulting Dataset
- 20. Date and Time features: year, month, week of month, etc. Feature Engineering: Create Data Features Season/Holiday features: New Year, U.S. Labor Day, U.S. Black Friday, and Christmas Fourier features to capture seasonality Lag features: these are values at prior time steps (weeks)
- 21. aka.ms/AIML30 #MSIgniteTheTour Demo We are using C# but you can do these steps in any language! @frlazzeri
- 22. C# Process Data: Add Weeks to Predict
- 23. C# Create Time Features: Extract Concept of Time Date and Time features: year, month, week of month, etc.
- 24. C# Create Holiday Features: Extract Holidays from Time Season/Holiday features: New Year, U.S. Labor Day, U.S. Black Friday, and Christmas.
- 25. C# Create Fourier Features: Capture Up/Down Pattern Fourier features to capture seasonality
- 26. C# Create Lag Features: Capture Prior Weeks to Current Lag features: these are values at prior time steps
- 29. aka.ms/AIML30 #MSIgniteTheTour Modeling Selecting the “Right” Algorithm to Train Your Model @frlazzeri
- 32. Deployment with Azure Machine Learning
- 33. Demo: Build, Test and Deploy Your Model
- 34. The Machine Learning Process @frlazzeri
- 35. Useful Resources: @frlazzeri ✓ AzureML GitHub: aka.ms/AzureMLrepo ✓ Algorithm Cheat Sheet: aka.ms/AlgorithmCheatSheet ✓ Deep Learning VS Machine Learning: aka.ms/DeepLearningVSMachineLearning ✓ Automated Machine Learning Documentation: aka.ms/AutomatedMLDocs ✓ Model Interpretability with Azure ML Service: aka.ms/AzureMLModelInterpretability ✓ Azure Machine Learning Service: aka.ms/AzureMLservice ✓ Azure Machine Learning Designer: aka.ms/AzureMLdesigner ✓ Get started with Azure ML: aka.ms/GetStartedAzureML ✓ Azure Notebooks: aka.ms/AzureNB
- 36. Coming next… ODSC East • Boston April 16th, 2020 • 14:00 PM Boston Hynes • Convention Center Training and Operationalizing Interpretable Machine Learning Models https://opendatascience.com/training-and-operationalizing-interpretable-machine-learning- models/ @frlazzeri