Introduction
- Usage of algorithms to derive predictive insights from data and make repeated decisions
- Scaling up business intelligence and decision making
- Forwards looking—compare with analytics (backwards looking)
- A computer writes a program for another computer
- Computer figures out best program by looking at examples
Vs AI
- AI
- Discipline (c.f. Physics)
- Abilities of computers to mimic human behaviours
- ML
- Methodology
- Generally refers to supervised and unsupervised learning
What Problems Can ML Solve?
- Things currently done via ‘rules’
- Scale, automate, personalise
- Not for problems were one set of rules fits all—generic
- Personalisation—lots of data required
Best Practices
Concepts and Definitions
Term | Definition |
---|
Example/instance/observation | Row in data set |
Label | Value to predict. Can be numeric (requires regression model) or categorical (requires logistic model) |
Features | Columns which aren’t labels. |
Feature engineering | Combine/transform features. |
One-hot encoding | Transform categorical data to numeric. 1 or 0 to indicate whether category applies to the observation |
Hyperparameters | User-defined parameter used to control the learning process. |
Model Types
- Supervised learning—task driven, identifies a goal, labelled data
- Classification—predict a category
- Regression—predict a continuous number
- Unsupervised learning—data-driven, identifies a pattern, unlabelled data
- Clustering
- Association—identify underlying relationships
- Dimensionality reduction—transform data from a high-dimensional space to a low-dimensional space, combine features
Model Architectures
- Convolutional Neural Networks (CNNs)—image classification, object detection, recommendation systems
- Recurrent Neural Networks (RNNs)—sequence modelling, next word prediction, translation, speech-to-text
- Sorting/clustering—anomaly detection, pattern recognition
- Generational Adversarial Networks (GANs)—anomaly detection, cyber security, self-driving cars, reinforced learning
Strategy
- Follow The Path to Machine Learning
- Starts as data strategy
- Start with manual data analysis
- Fail fast, try new ideas in an agile way
- Then move to ML
- Data collection is the longest/hardest part—most likely to fail
- If you can’t do analytics, you can’t do ML
ML Effort Allocation
- Expectation:
- Optimizing ML algorithm is most significant effort—practitioners focus on latest ‘shiny’ thing
- Reality:
- Data collection and building infrastructure often overlooked
- By spending time collecting data, KPIs often become apparent—rather than relying on intuition