Purest row

The first step in analysing a data table is to determine how many pure factors have to be estimated. Basically, there are two approaches which we recommend. One starts with a PCA or else either with OPA or SIMPLISMA. PCA yields the number of factors and the significant principal components, which are abstract factors. OPA yields the number of factors and the purest rows (or columns) (factors) in the data table. If we suspect a certain order in the spectra, we preferentially apply evolutionary techniques such as FSWEFA or HELP to detect pure zones, or zones with two or more components. 

Depending on the way the analysis was started, either the abstract factors found by a PCA or the purest rows found by OPA, should be transformed into pure factors. If no constraints can be formulated on the pure factors, the purest rows... 

There are many chemometric methods to build initial estimates some are particularly suitable when the data consists of the evolutionary profiles of a process, such as evolving factor analysis (see Figure 11.4b in Section 11.3) [27, 28, 51], whereas some others mathematically select the purest rows or the purest columns of the data matrix as initial profiles. Of the latter approach, key-set factor analysis (KSFA) [52] works in the FA abstract domain, and other procedures, such as the simple-to-use interactive self-modeling analysis (SIMPLISMA) [53] and the orthogonal projection approach (OPA) [54], work with the real variables in the data set to select rows of purest variables or columns of purest spectra, that are most dissimilar to each other. In these latter two methods, the profiles are selected sequentially so that any new profile included in the estimate is the most uncorrelated to all of the previously selected ones. 

In Section 34.2 we explained that factor analysis consists of a rotation of the principal components of the data matrix under certain constraints. When the objects in the data matrix are ordered, i.e. the compounds are present in certain row-windows, then the rotation matrix can be calculated in a straightforward way. For non-ordered spectra with three or less components, solution bands for the pure factors are obtained by curve resolution, which starts with looking for the purest spectra (i.e. rows) in the data matrix. In this section we discuss the VARDIA [27,28] technique which yields clusters of pure variables (columns), for a certain pure factor. 

The rows with the highest purities are estimates of the row factors, i.e. the purest spectra from the data set, which are refined afterward by alternating regression. 

The rows related to the purest pixels will provide good approximations of the pure spectra sought, whereas the columns linked to the purest spectral channels will allow for building approximate distribution maps of the pure constituents. 

---