Hi Jonathan, thanks for your question. The following was meant: For i=1,2,4 build a classifier consisting of: 1. Transformation of the full Iris dataset to i-dimensions using PCA (do not use label information here). For i=4 this has no effect obviously. 2. Two linear SVM trained on the i-dimensional data, for example as you described. A label is then computed by plugging the i-dimensional representation obtained from PCA into the two SVMs in a suitably way. We want to check how the accuracy relates to i.
For the actual classifier itself, the data is projected with with the red PCA matrix and the green PCA matrix. These processed data are given as input to the red and green SVMs respectively (in a sequential manner). Then the result is constructed in the obvious way based on the SVM results.
However, I am particularly confused by the comment “print accuracy on 1, 2, 4 dimensions”, for based on my understanding, we are supposed to use at least two different dimensions in our classifier.
I do not get your understanding of the task unfortunately. The first sentence reads »[…] build classifiers for the whole Iris dataset which use a 4D, 2D, or 1D PCA respectively as the first step, […]«. Maybe you could elaborate in order to improve the wording on the sheet? Would »[…] build three independent classifiers […]« help? Probably the term »classifier« is understood as referring to the subsequent SVMs and it is not clear that a classifier shall consist of 1 PCA + 2 SVM with a total of three classifiers? If you have problems you can also come to the tutorial on Monday to ask Lisa. Kind regards, Jannik Schürg
Jonathan _______________________________________________ Mllab mailing list Mllab@ins.uni-bonn.de https://mail.ins.uni-bonn.de/mailman/listinfo/mllab