Please enable JavaScript.

Coggle requires JavaScript to display documents.

Similarity, Neighbors, Clusters (Nearest-Neighbor reasoning (Issues…

- - - - Scan through training examples and choose several that are most similar to the new example
  - - - Lack of explicit interpretable model
      - Justification of specific decision
      - Of entire model
    - - Too many attributes
      - Too many irrelevant attributes
      - Way to fix
        
        Feature selection
        
        Judicious determination of features
        
        Either manually by background knowledge, or by data mining automatically
  - - - Training fast because only storing the instances
      - Most of cost comes from prediction/classification step
        
        Can be expensive
  - - - More attributes, more complex
        
        Example: categorical and numeric data, can't use Euclidean distance formula
        
        Can us Manhattan distance formula
        
        Jaccard distance formula
        
        Cosine distance
        
        Particularly useful when want to ignore differences in scale
- - - - For a variety of different values of k
        
        Search for one that gives best performance
    - - Usually has tools to set K automatically
- - - - where
        the objects within groups are similar, but the objects in different groups are not so
        similar.
    - - For data rich companies
  - - - This structure is why it is called Hierarchical Clustering