| Author | Message | ||
     davrieux fabrice (Fabrice) |
Bonjour, (in french in the texte) I need some help in calculation of H value for supplementary sample. If I had no problem to calculate H (normalised) values for training set samples I did have trouble to estimate H value of a new sample projected on the previously calculated axes (PCs). In fact I don�t match the Winisi 1.5 results. I used the Mahalanobis formula: d²(g,i) = (CPi1)²/v(CP1) + (CPi2)²/v(CP2) for two PCs for example. Where g is the origin and v the variance of the PC (i.e the eigen values), and PCi. the scores of this new sample on the pcs. In fact I am not so far of ISI value if I don�t divide by the H mean value! Certainly I misunderstood something, and I do the wrong calculation, so I will be very pleased if you can help me. Thanks (merci infiniment) | ||
| Tony Davies (Td) |
Dear Fabrice, I asked Tom Fearn if he could help. Here is his answer: I think this is an ISI question. The formula (in so far as I can read it) looks right. The obvious source of problems is not calculating the scores correctly, eg forgetting to pretreat the new data (centering? scaling?) in exactly the same way as the training data before calculating the scores. Tom So the advice is ask ISI. Hope you get there! Best wishes Tony |