Lesson 1, 2: AWS Machine Learning Scholarship Program

Components of Machine Learning

1. Machine Learning Model

- Generic porgram, made specific by data

- A block of code used to solve different problems 

- A linear regression model example: to predict number of people attending college

 

2. Model Training Algorithm

- An iterative process 

- Determine what changes need to be made to get to the end goal

- Makes small changes to the model to make these results better

 

3. Model Inference Algorithm

- Process to use a trained model to solve a task

- Using a trained model to generate predictions: model inference

 

Major steps in the ML process

1) Define the problem

2) Build the dataset

3) Train the model

4) Evaluate the model

5) Use the model

 

1. Define the problem

• Be specific
• Identify the ML task we might use to solve the problem 
• Two main ML tasks: Supervised & Unsupervised learning

Supervised tasks use labeled data wheras unsupervised tasks use unlabeled data.

Classificaiton tasks: predicting some unknown categorial attribute about your data

Regression tasks: predicting some unkown continuous attribute about your data

Clustering tasks: exploring how your data might be grouped together 

 

2. Build a dataset

4 steps

• Data collection: Find and collect data related to the problem you have defined.
• Data inspection: Explore your dataset looking for outliers, missing or incomplete data and dataset needs to be transformed
• Summary statistics: Summary statistics can identify trends in/scale of/shape of the data
• Data visualization: Great data visualizations communicate the findings to project stakeholders 

3. Training a model

Splitting Dataset

- Majority will be held in the training dataset (80%)

- The test dataset (20%) will be used during model evaluation

 

Model Training Terminology

Literatively update model parameters to minimize some loss funciton

- Model parameters: Configuration that changes how the model behaves 

- Loss function: Measurement of how close the model is to its goal

 

Putting it all together

- Feed the training data into the model

- Compupte the loss function on the results

- Update the model parameters in a direction that reduces loss

 

A few other details...

- How do I actually implement model training?

- How do I determine which model to use? (model selection, may try different types of models while solving a problem with ML)

- Training algorithm hyperparameters (settings on the model which are not changed during training such as the number of clusteres the model should identify)

- Be prepared to iterate

 

Different model types

• Linear models
• Tree-based models
• Deep learning models: neural network (the model) that is composed of collections of neurons connected together by weights (mathematical representations of how much information to allow to folow from one neuron to the next): examples include FFNN, CNN, RNN/LSTM, Transformer 

4. Model Evaluation

Using model accuracy 

Using log loss

- calculate how uncertain your model is about the predictions it is generating

Exmaple) Creating a model for predicting orthopedic pathology using Amazon SageMaker 

 

5. Model inference

Use your model to solve real problems 

Monitor the results