liveBook · Manning

Topographic error

As explained in Build AI Drug Discovery Pipelines

Topographic error is the proportion of data points whose first and second best matching units are not adjacent in a self-organizing map.

Topographic error measures how well a self-organizing map preserves neighborhood structure. For each data point, identify its first and second best matching units (BMUs), then check whether those units are adjacent. The metric is the proportion for which they are not; lower values indicate better topology preservation. section 7.1.2

Use topographic error to compare SOM configurations:

  • Hold out 20% of the compounds while training and evaluating candidate maps. section 7.1.2
  • Vary neighborhood width, learning rate, and distance function, then select the configuration with the lowest topographic error. section 7.1.2
  • Interpret the metric alongside quantization error: topographic error assesses organization, while quantization error measures distance from each point to its BMU. section 7.1.2

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines the BMU-adjacency calculation and applies it during held-out parameter evaluation, without deriving the SOM update algorithm.
Examples
Evaluates maps of 2,062 compounds represented by 2,048-dimensional Morgan fingerprints using MiniSom.
What it assumes
Assumes familiarity with data points and distance functions while defining BMUs and topology preservation operationally.
What you can do after
Evaluate and select a self-organizing map using held-out topographic error.
How it connects
Connects self-organizing maps, best matching units, quantization error, Morgan fingerprints, and cosine distance.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.