Generalization
- Algorithms use example data to create a generalization (a model) that answers your business question
- After creating a model using example data, use it to answer your business question with new data (obtaining an inference)
Understand the Business Problem
The type of answer sought influences your choice of algorithm.
Discrete Categories
Example: Direct mail campaign to solicit political donations
Type answers
- “Considering past solicitation responses, mail to this contributor ?”
- Binary:
yes/no
- Binary:
- “Considering past solicitor segmentation, which segment is the contributor in ?”
- Multi-class:
small donation
,significant donation
,corporate donor
- Multi-class:
SageMaker built-in algorithms
- Linear Learner
set
predictor_type
hyperparameter tobinary_classifier
- XGBoost
set
objective
hyperparameter toreg:logistic
Quantitative
Example: Direct mail campaign to solicit political donations
Type answers
- “Considering the ROI on past solicitation mailings, what is the ROI for soliciting this donor ?”
- Quantitative:
higher ROI donors get mailing
- Quantitative:
SageMaker built-in algorithms
- Linear Learner
set
predictor_type
hyperparameter toregressor
- XGBoost
set
objective
hyperparameter toreg:linear
Discrete Recommendations
Example: Direct mail campaign to solicit political donations
Type answers
- “Considering past solicitation mailing responses, what is the recommended content for each donor ?”
SageMaker built-in algorithms
- Factorization Machines for recommendations
Identifying Groups
Example: Direct mail campaign to solicit political donations
Type answers
-
“Group potential and current donors into 12 groups based on their attributes, how should they be grouped ?”
-
Send mailing to donors in the group that has the highest percentage of current donors, i.e. potential donors that are most similar to current donors
SageMaker built-in algorithms
- K-Means
Dimensionality Reduction
Example: Direct mail campaign to solicit political donations
Type answers
-
“What attributes differentiate donors, what are the relative values for the donors along these dimensions ?”
-
Use to simplify the view of current and prospective donors and gain clarity on value of donor attributes
SageMaker built-in algorithms
- Principal Component Analysis
Other algorithms
- Classifying images
- Image Classification
- Translation
- Sequence-to-Sequence
- Determining topics of sets of documents
- Latent Dirichlet Allocation
- Neural Topic Model
- Text classification
- Blazing Text
- Anomaly detection
- Random Cut Forest
- k-Nearest-Neighbors
本文由
Oscaner
创作, 采用
知识共享署名4.0
国际许可协议进行许可
本站文章除注明转载/出处外, 均为本站原创或翻译, 转载前请务必署名