AWS Machine Learning with SageMaker: Hands-On

Following Downloadable Resources are available in this lecture:

1. Source Code Setup Document

2. Introduction to Machine Learning and Concepts Document

3. usa_airpassengers_numeric.xlsx

AmazonSageMakerCourse Git Repo:
https://github.com/ChandraLingam/AmazonSageMakerCourse.git

The attached Model Performance Evaluation.pdf (which is part of this lecture) contains the important formula for metrics calculation

"The XGBoost (eXtreme Gradient Boosting) is a popular and efficient open-source implementation of the gradient boosted trees algorithm". 

https://docs.aws.amazon.com/sagemaker/latest/dg/xgboost.html

Introduction to XGBoost and how it compares to the Linear Model, Decision Tree, and Ensemble Methods

Let's compare performance of XGBoost and Linear Model using a simple regression dataset

Train model using XGBoost and Linear Regression. Evaluate performance using Plots, Residual Histograms and RMSE metrics

Let's compare performance of XGBoost and Linear Model using a non-linear data set

In this lab, let's look at Bike Rental demand forecasting problem.  This is an old competition problem from Kaggle: https://www.kaggle.com/c/bike-sharing-demand/data.  To download data files, you need to register with Kaggle (it's free).

In this lab, let's train our model for forecasting hourly bike rental counts.  This is a complex non-linear data set that has seasonality, trend and several factors that impact rentals.  Evaluate quality of predictions using Plots, Residual Histograms, RMSE and RMSLE metrics.  Finally, submit the results at Kaggle for test data.

In this lab, let's transform the target using log operation.  Log of target can help when the target is a count/integer, it has seasonality and trend.  After model predicts the value, we need to apply inverse transform (exp) to get the count back.

In this lab, let's train bike rental model on SageMaker's built-in XGBoost Algorithm.  We will walk through the fours steps for using a SageMaker algorithm

In this lab, let's look at the steps involved in connecting to an existing SageMaker endpoint, security of an endpoint, how to send multiple observations in each call.

Let's look at key benefits of a managed Endpoint.  SageMaker takes care of automatic replacement of unhealthy instances, AutoScaling infrastructure based on workload, hosting multiple versions of model behind an endpoint, and metrics published to CloudWatch.

In this lab, let's look at multi-class classification using XGBoost.

In this lab, let's look at a how to perform Binary Classification using XGBoost.  We will use the diabetes data set in this lab

In this lecture, let's look at important XBoost Hyperparameters.  We will also look at Bias, Variance, Regularization (L1, L2), and Automatic Tuning