*** NEW PREVIEW VIDEOS: Take a look at several newly enabled Preview videos. All lectures in Section 3 and Section 4 on Linear Regression are available for preview as well as Section 15 Integration objectives
Note: AWS Machine Learning is not part of free-tier. So, you will incur a small charge when creating and running prediction on models. For this course, I spent USD 5-6 total for creating and testing all models. ***
This course is designed to make you an expert in AWS Machine Learning and it teaches you how to convert your cool ideas into highly scalable products in a matter of days.
Biggest challenge for a Data Science professional is how to convert the proof-of-concept models into actual products that your customers can use. There are several courses on machine learning that teach you how to build models in R, Python, Matlab and so forth. However, converting a model into a scalable solution and integrating with your existing application requires a lot of effort and development. The real success of your ideas and concepts depends on how soon you can put the capabilities in the hands of your customers.
With AWS Machine Learning service, you can easily conduct experiments and test your concepts. Once you are happy, you can instantly scale to support millions of requests. No separate development work needed.
This course is focused on three aspects:
The Core of the machine learning process is the algorithm itself. Gaining an intuitive understanding of the algorithm, how does it find the solution, and what are the knobs to tweak is essential for a successful career in this field. That is where we will focus first.
Once we build the model, how do we know if it is good or bad? Or If we want to compare two different models, how do we decide which one to pick? We will look at industry standard metrics and powerful visualization tools that AWS provides to assess the goodness of a model.
The third aspect and most exciting part of model development is putting the prediction capability in the hands of the users, validate how they are using it and identify what needs to be refined. There is a whole section in this course dedicated to integration of machine learning models with your application. We will walk thru several integration and security options.
This course is completely hands-on with examples using: AWS Web Console, Python Notebook Files, and Web clients built on AngularJS. You will also learn and integrate security into exercises using variety of AWS provided capabilities including Cognito.
There are Quizzes and supporting resources as well.
Introduction to AWS Machine Learning Course, Topics Covered, Course Structure
1. Setup Course Folder in local machine
2. Download Project Source Code
3. Download Data files
Introduction to Python Development Environment, Pandas, NumPy, Matplotlib
Setup your AWS Root Account
Configure Identity and Access Management Security. Add lower privilege IAM user accounts that will be used for hands-on machine learning exercises.
Summary of Introduction, Development Environment Setup and AWS Configuration
Learn about standard machine learning terminologies that we will be using in this course
Introduces data types supported by AWS Machine Learning with examples
Introduces linear regression models with examples; hands-on python data analysis for assessing the input features.
Introduces binary classification models with examples; hands-on python data analysis for assessing the input features.
Introduces multiclass classification models with examples; hands-on python data analysis for assessing the input features.
Overview of plotting capability in python using Matplotlib and how various functions appear in a graph. Useful for understanding relation between input and output attributes.
Understand concepts behind linear regression. How does the algorithm predict a numeric value, what optimization steps does it go through, how does it evaluate cost/loss and more
Understand how linear regression is able to fit non-linear shapes
Create Training Data and Characteristics of input features and target attribute
How to create datasource in AWS, how to review statistics, missing values, invalid values, schema definition
How to create a new model in AWS with default configuration recipe
Review RMSE, Residual Histograms for the model we trained in AWS
Review the configuration or recipe used for training the model
Learn how to customize the recipe or configuration for training a model
Review the results of default recipe versus custom recipe models
Summarize the learnings
Prepare training data with quadratic features
How would a model respond to a condition where there are lot more features than necessary?
What would happen if magnitude of features are very different?
Let's find out!
How to fix a condition where features with large magnitude dominate the outcome?
Normalization to the rescue!
Let's rebuild the model with normalization enabled through recipes.
How is the model predictive quality now?
Having lot of features and higher order generally help improve the model quality. However, when the magnitudes are very different, consider normalizing the numeric features
Prepare a training data target features having a complex shape
Learn how to add complex features to your training data
Train model with variety of input feature scenarios
Assess model performance that fits complex shape with degree 1 features
Assess model performance that fits complex shape with degree 4 features
Assess model performance that fits complex shape with degree 15 features
Summary of performance findings and how higher order features enable fitting complex shape
Review Problem, Initial Data Assessment, Features, Data Type and Model Strategy
Assess Prediction Quality, Performance Summary and Next Steps
Linear Regression Summary and Concepts we have covered so far
Learn about classification problem, solution objectives, model representation, probability score and cut-off
Simple example that shows how logistic regression process works
How to evaluate performance of a classification algorithm. Terminology introduction
Discuss objectives of the optimization algorithm
Logistic cost function characteristics
Exercise to understand how differences between actual and predicted value affect cost
How does algorithm march towards optimal solution
Summary of logistic regression process, logistic function, cost function, optimization algorithm and terminologies
Discuss problem, solution objective, input features, target attribute and assessment of data
Prepare Training, Evaluation and Test Datasets
Training a Classification model on AWS
Metrics to evaluate predictive quality of a classification algorithm. True Positive Rate, False Positive Rate, Accuracy, Precision and more
Visual Insight into predictive quality of model using powerful Positive and Negative Histograms provided by AWS
Area Under Curve Metrics for summarizing prediction quality
Assess quality of Diabetes prediction model using AUC, Positive Negative Histograms and other metrics
Use new dataset and assess model performance using AWS Model Evaluation capability
Summary of Binary Classifiers, Training Models and Predictive Quality Metrics
Chandra Lingam spent 15 years at Intel, developing and managing systems that handled hundreds of terabytes of worldwide factory data. Chandra is an expert on Amazon Web Services, mission critical systems and machine learning. He has a Master's degree in Computer Science from ASU and Bachelor's degree in Computer Science from Thiagarajar College of Engineering, Madurai.
Chandra is the author of popular iOS educational apps Geometry Test, Math Stripes and Arithmetic Test.