Dimensionality reduction in infrared microscopy is aiming to preserve both spatial and molecular information in a more compact manner and therefore reduces the computational time for subsequent classification or segmentation tasks. We offer four different dimensionality reduction approaches mentioned in Dimensionality reduction for deep learning in infrared microscopy: A comparative computational survey, namely Principle Component Analysis, Uniform Manifold Approximation and Projection and two different Contractive Autoencoder.
This repository contains four different scripts:
- PCA_tf2.py : Performs principle component analysis on spectral data of shape (X*Y,s pectra) with n_components. Data can be standarized such that they have a mean of 0 and standard deviation of 1.
- UMAP_tf2.py : Performs uniform manifold approximation and projection on data of shape (X*Y, spectra) with n_neighbors, n_components, min_dist and metric. Data can be standarized such that they have a mean of 0 and standard deviation of 1.
- FCCAE_tf2.py: Training of the entire stacked autoencoder from scratch, yielding a fully connected contractive autoencoder with Tensorflow.
- SCAE_tf2.py : Training of a series of stacked contractive autoencoders which are trained with one hidden layer each and are afterwards connected to form a deep autoencoder with Tensorflow.
Learn more here.
Prof. Dr. Axel Mosig: Bioinformatics, Center for Protein Diagnostics (ProDi), Ruhr-University Bochum, Bochum, Germany