Self-organising attribute maps and pattern spectra

This project combines Max tree data structure (implemented by DISCCOFAN and DISCCOFAN-SOM) and Self Organising Maps (implemented by minisom) to develop self-organising attribute maps and pattern spectra. This tool explores morphological structures in images.

To use the treeSOM code, you need to install DISCCOFAN (disccofan-master and disccofan-SOM for filtering based on the SOM) and minisom first, following the instructions in the links above.

How to use treeSOM?

First, create a work directory to store your data (images) and results. Your training images should be stored in image_directory. tree_directory corresponds to the directory where max trees of the training images will be stored. (You can also directly put your tree files into tree_directory)

base_directory = '/project/treeSOM/'
image_directory = base_directory + '/images/'
tree_directory = base_directory + '/trees/'
disccofan_directory = '/project/software/disccofan-master/'
disccofanSOM_directory = '/project/software/disccofan-SOM/'
     

Load image files to train.

imagefiles = TreeSOM.LoadImageDirectory(image_directory, extension='fits')
     

Plot training images all at once.

TreeSOM.ShowImages_in_Parallel(
    imagefiles,
    base_directory,
    cmap = 'inferno',
)
     

Use ShowImages_in_Parallel instead, if the number of training images is larger than 25.

Create max trees for the training images.

TreeSOM.CreateTrees_in_Parallel(imagefiles,
    'fits',
    disccofan_directory,
    base_directory,
    tree_directory, )
     

Load (sampled) vector attributes from nodes of the max trees. tree_selecting_rate indicates the fraction of trees to load from the entire data sets and sampling_rate indicates the fraction of nodes to load within each tree.

train_set = TreeSOM.LoadTrees(
    base_directory,
    tree_directory,
    extension = "csv", 
    tree_selecting_rate = 0.1,
    sampling_rate = 0.005,   )

Set parameters for the self-organising map training and train a SOM.

n_neurons = 15
m_neurons = 15
s_features = [4,5,6,8,9,13,14]
max_iter = 100000

TreeSOM.TrainSOM(train_set, base_directory, n_neurons, m_neurons, s_features, max_iter,)