ugtm 2.3.0

Generative Topographic Mapping (GTM) for Python: GTM classification and regression

GTM (Generative Topographic Mapping) is a dimensionality reduction algorithm (like t-SNE, LLE, etc.) created by Bishop et al. and a probabilistic counterpart of Kohonen maps. ugtm implements GTM and GTM-based prediction algorithms, including a kernel variant (kGTM), classification (GTC) and regression (GTR) maps, sklearn-compatible estimators, and repeated cross-validation.

Full documentation: https://ugtm.readthedocs.io/

Quick start

import ugtm
import numpy as np

data   = np.random.randn(100, 50)
labels = np.random.choice([1, 2], size=100)

gtm = ugtm.runGTM(data=data)

coordinates = gtm.matMeans   # mean positions  (n_samples, 2)
modes       = gtm.matModes   # mode positions  (n_samples, 2)
resp        = gtm.matR       # responsibilities (n_samples, n_nodes)

sklearn-compatible estimators

from ugtm import eGTM, eGTC, eGTR

transformed      = eGTM().fit(X_train).transform(X_test)
predicted_labels = eGTC().fit(X_train, y_train).predict(X_test)
predicted_values = eGTR().fit(X_train, y_train).predict(X_test)

Large datasets: incremental GTM (iGTM)

For datasets too large to hold the full N×K responsibility matrix in RAM, use iGTM (Gaspar et al. 2014). Data is processed in blocks; only two small accumulators are kept per iteration instead of the full N×K matrix:

from ugtm import runIGTM, eIGTM

# Low-level wrapper — same interface as runGTM
model = runIGTM(data, n_blocks=10)
coordinates = model.matMeans   # (n_samples, 2)

# sklearn transformer — n_blocks=0 chooses block size automatically
transformed = eIGTM().fit(X_train).transform(X_test)

# Block-wise projection for large test sets (generator, bounded memory)
for block in eIGTM().fit(X_train).transform_blocks(X_test, block_size=1000):
    pass  # process each (block_size, 2) chunk here

Visualisation

ugtm outputs are plain NumPy arrays — use any plotting library:

import matplotlib.pyplot as plt

gtm    = ugtm.runGTM(data=data)
coords = gtm.matMeans

plt.scatter(coords[:, 0], coords[:, 1], c=labels, cmap="Spectral_r")
plt.colorbar()
plt.show()

See https://ugtm.readthedocs.io/ for richer examples.

Predictions and cross-validation

# GTM classification / regression
predicted = ugtm.GTC(train=train, test=test, labels=labels)
predicted = ugtm.GTR(train=train, test=test, labels=activity)

# Repeated cross-validation
ugtm.crossvalidateGTC(data=train, labels=labels, s=1, regul=1)
ugtm.crossvalidateGTR(data=train, labels=activity, s=1, regul=1)

References

1. GTM algorithm — Bishop et al. (1998)

2. Kernel GTM — https://www.elen.ucl.ac.be/Proceedings/esann/esannpdf/es2010-44.pdf

3. GTM classification models — https://www.ncbi.nlm.nih.gov/pubmed/24320683

4. GTM regression models — https://www.ncbi.nlm.nih.gov/pubmed/27490381

5. ugtm paper — https://openresearchsoftware.metajnl.com/articles/10.5334/jors.235/

6. Incremental GTM — Gaspar et al. (2014), Chemical Data Visualization and Analysis with Incremental GTM: Big Data Challenge