API Reference¶

The public API is small by design. Most workflows use the SOM class, the SOMVisualizer, and (optionally) the SOMConfig. The pages below document every module in full.

Core

torchsom.core — the SOM class and its BaseSOM interface.

Core API

Utils

torchsom.utils — distances, neighborhoods, decay, clustering, metrics, search.

Utilities API

Visualization

torchsom.visualization — SOMVisualizer and VisualizationConfig.

Visualization API

Configs

torchsom.configs — the Pydantic SOMConfig model.

Configuration API

Public objects at a glance¶

Object	Purpose
`torchsom.core.SOM`	The Self-Organizing Map: `fit`, `build_map`, `cluster`, `collect_samples`.
`torchsom.core.BaseSOM`	Abstract base defining the SOM interface and shared attributes.
`torchsom.visualization.SOMVisualizer`	Factory that renders every plot for the SOM’s topology.
`torchsom.visualization.VisualizationConfig`	Styling options for figures (size, fonts, colormap, DPI, hex settings).
`torchsom.configs.SOMConfig`	Validated, serializable configuration for a `SOM`.
`torchsom.DISTANCE_FUNCTIONS`	Registry of distance metrics (`euclidean`, `cosine`, `manhattan`, `chebyshev`).
`torchsom.NEIGHBORHOOD_FUNCTIONS`	Registry of neighborhood kernels (`gaussian`, `mexican_hat`, `bubble`, `triangle`).
`torchsom.DECAY_FUNCTIONS`	Registry of learning-rate and neighborhood-width decay schedules.

Import conventions¶

The most common objects are re-exported at the top level:

from torchsom import SOM, SOMVisualizer
from torchsom.visualization import VisualizationConfig
from torchsom.configs import SOMConfig

The function registries let you list or extend the available options:

from torchsom import DISTANCE_FUNCTIONS, NEIGHBORHOOD_FUNCTIONS, DECAY_FUNCTIONS

print(list(DISTANCE_FUNCTIONS))       # available distance metrics
print(list(NEIGHBORHOOD_FUNCTIONS))   # available neighborhood kernels