You will be able to:

•  
• Evaluate multiple supervised classification algorithms using a variety of advanced techniques
• Understand and use a variety of specialized supervised learning algorithms
• Understand and use algorithm boosting
• Learn how to engineer new data features to increase model power
• Understand and use model bagging
• Perform label creation and analysis for proper threshold determination
• Appropriately deal with imbalanced (rare events) and small datasets
• Determine the most powerful predictors using multiple techniques
• Know multiple ways to transform your data to optimize model predictive power
• Know how to transform time-series data for supervised learning
• Use AWR and V$ Sysmetric data in your ML models
•  

Class Pricing:

• $449 USD for Free members
• $359 USD for Essential and Advantage members (ask for your coupon code)
• FREE for Prime members (just ask to be registered)

Class Outline:

Session 1. Core Supervised Machine Learning

• Core Predictive ML
  • Review unsupervised learning
  • Supervised learning
  • Choosing between classification and regression
  • Build and predict flows
  • Features
  • Labels - classic
  • Time series prediction
  • Labels - time series classification using percentiles
• Decision Trees Introduction
  • Classic decision trees
• Bagging Decision Trees
  • Introduction to bootstrap aggregation
  • Explore various types of bagged decision trees
• Discover The Best Algorithm... Fast!
  • General strategy
  • Simple scoring
  • Reduce dataset by random sampling
  • Reduce model evaluation intensity
  • Quickly build and evaluate multiple models

Session 2. Advanced Preprocessing & Evaluation

• Random Forest Algorithm Introduction
• Advanced Preprocessing
  • Feature engineering: power (X^y)
  • Feature selection strategies
• Advanced Evaluation Techniques
  • Why simple scoring is disastrous
  • Imbalanced datasets
  • Testing using different data
  • Time series repeated k-fold evaluation
  • Using the Confusion Matrix
  • Using F1 score
• Quickly Building And Evaluating Multiple Models
• How To Improve Model Power

Session 3. Rare Event Datasets & Boosting Algorithms

• Working With Imbalanced (rare event) Datasets
  • Strategies and approaches
  • Detection
  • Algorithm parameters: balance, class weights
  • Oversampling using SMOTE and ADASYN
• Model Boosting Techniques
  • General strategies
  • AdaBoost
  • XGBoost
• Suggest Steps For Continued Learning
• Quickly building and evaluating multiple models
  • Conda environment setup
  • Strategies to get insights fast
  • The entire process outlined
  • Quickly building and evaluating multiple models
• Realtime Prediction Simulator
  • Conda environment setup
  • Prediction preparations are different
  • The realtime prediction simulator

Details: 

Class Hours: There are 6 hours of live, interactive hands-on class training divided into three 2-hour sessions.

Class Structure: OraPub believes you learn more when you freely and actively engage during the training. This is why each session is full of live demonstrations, lots of Q&A time and lots of hands-on action! If you miss a session, you are free to attend that session when the class is offered again.

Watching Recorded Sessions: Each session is recorded and used to create a unique web page just for your class. The recordings can be streamed using any device, but not downloaded. You can re-watch the videos as many times as you wish. The videos and your unique class web page will be available for three years!

Class Requirements:

You will need:

• To have completed OraPub's Machine Learning 1 Anomaly Detection LVC. We assume you have mastered the ML concepts and have the hands-on experience that naturally results from our ML 1 LVC.
• Experience with any programming language, such as PL/SQL. If you are familiar with Python, that will be helpful, but in no way is it required. Craig has built an interactive "Python Primer" to quickly introduce Oracle professionals to Python. You can download the interactive Python Primer here: https://www.orapub.com/members > Tools - Presentations > Tools > Python Primer
• The ability to connect to the online training over the internet, hosted by Zoom. You do not need a microphone. Craig Shallahamer (Viscosity's Applied AI Scientist), your teacher, is usually the only one that speaks, but all students are able to type their questions, answers, and comments... it can be very lively and a lot of fun!
• A Gmail account with the ability to connect to the internet using the Chrome web browser. All your hands-on work will be done in Google's free Colabratory web-based environment.