1. Introduction to Machine Learning Flashcards
(25 cards)
What are some examples of questions that descriptive models can answer
What was the rate of returning customers buying on the website?
How many cases of fraud were investigated last month?
What were the email open, click-through, and response rates?
How many customers made a purchase after 5 minutes?
What are some examples of questions that predictive models can answer
What is the likelihood that the customer buys online again?
What is the likelihood that the transaction is fradulent?
What is the likelihood an email will be opened?
What is the likelihood the website visitor is an impulsive buyer?
What are the 3 types of analytics
- descriptive
- predictive
- prescriptive
What questions, enablers and outcomes are related to descriptive modeling
What happened? What is hapenning?
Business reporting. Dashboards. Scorecards. Data warehousing
Well-defined business problems and opportunities
What questions, enablers and outcomes are related to predictive modeling
What will happen? Why will it happen?
Data Mining. Text mining. Web/Media mining. Machine Learning
Accurate projections of future events and outcomes
What questions, enablers and outcomes are related to prescriptive modeling
What should I do? Why should i do it?
Optimization. Simulation. Decision modeling. Expert systems
Best possible business decisions and actions
O que significa OCR
Optimal Characther Recognition
(escrever à mão e passar automaticamente para texto)
Como funciona o OCR
Criar um dataset com imagens de letras manuscritas.
Cada imagem é rotulada com a letra correspondente (por exemplo, “A”, “B”, “C”…).
O modelo aprende a reconhecer padrões nas imagens para associar à letra correta.
What is a task
A task corresponde a uma atividade que se pretende que um sistema de ML execute, representa o objetivo funcional do sistema/aquilo que se deve aprender
O que é a experience
Experience refere se ao conjunto de dados ou interações que o sistem utiliza para aprender. Corresponde à fonte de informação
O que é a performance measure
Métrica usada para avaliar o quao bem o sistema está a realizar a task
No caso de OCR, a que corresponde a task, experience e a measure
Task = hand right letter recognition
Experience = set of hand written letters labeled by humans
Performance measure = percentage of letters classified correctly
Assuming we have email program that based on emails that the user marks as spam or non spam, learns how to improve its own spam classification. What is a task (T), Experience (T) and Measure (M)?
Task = application that classifies emails as “spam” or “not spam”
Experience = see how the user marked the email
Performance = the number or fraction of emails correctly classified
What are the 3 types of Machine Learning problems
- Supervised Learning (Regression and Classification)
- Unsupervised learning (clustering and dimension reduction)
- Reinforcement learning
1 and 2 are semi-supervised learning
What is supervised learning
Uses labeled inputs attributes to predict an outcome
what is a classification model
process of finding a model to predict data classes or concepts (the outcome is categorical)
what is a regression model
process of finding, a model to predict numeric outcomes (the outcome is continuous)
What is unsupervised Learning
when the input are not labeled and there is no target
what are some examples of supervised learning
regression and classification
what are some examples of unsupervised learning
clustering e dimension reduction
what is semi-supervised learning
makes use or non-labeled input attributes to gain more understanding of the population
what is reinforcement learning
when inputs attributes are not labeled, or labels are not defined, and model learns from a rewarding process
what are some benefits of using predictive models in marketing
- Develop and manage customer relationships
- Maximize customer lifetime value / share of value
- customers are your assets
- structure and manage organization around customer segments
- user tecnologies and configure processses to find ways to customize interactions
- Use targeted distribution and media
