This workbook is an attempt to reproduce some unpublished research which I undertook in a previous life when working for the Shidmadzu Research Laboratory in the late '90s. Although similar research has subsequently been published, as far as I am aware, this was original research at the time. I used this as a learning exercise for parts of Pandas, NumPy and sklearn.The context of this research is the visualisation of multi- or hyper-spectral images, where each pixel of an image consists of a spectrum. Originally conceived as part of the LandSat programme, multi-spectral imaging was taken into the field of surface analytics by my PhD supervisor, the late Prof. Martin Prutton.Visualisation of multi-spectral data presents a significant challenge, not least because of the need to be able to visualise data in considerably more than three dimensions. My original approach was to quantise the data using a Self-Organising Map (SOM), and then use a Sammon Mapping to organise the map vectors into a two-dimensional space where the distance between the vectors in the reduced data space approximates their (Euclidian) distances in the original data space. This two-dimensional mapping can be transferred to two of the three dimensions of a colour space in order to visualise the original data.We start with some imports etc.

MattWenham

Multivariate Data Visualisation using Self-Organising Maps