Address feedback from Dan

[x] Preprocessing: Overview first -- what are the steps, input/output
[x] Before background: Overview of what we do (what is the problem). "What the data looks like etc to give the reader a flavour of what this is about"
[x] Rephrase Abstract (what is this about, what is the goal?)
[x] Move Introduction to ML, should be before PCA
[x] Domain-specific features histogram: Add legend
[x] References for techniques at "To avoid overfitting, ..."
[x] Reference for quadratic programming
[x] Figure: Reconstructions with various k-values: Add axis
[x] Figure of components: Explain exactly what this is (vectors from the U matrix)
[x] Running PCA: Make it clear that the covariance matrix is based on mean-normalized data (x hat)
[x] Replace (u, s, v) = svd(sigma) with A = U D Vtransposed
[x] PCA/Reconstruction: Formula, what is done? Moving from one space to the other. B = ux + my (ish)
[x] PCA/Choosing k: Relation between variance retained and eigenvalues ("the eigenvalue is the variance along this eigenvector")
[x] SVM intro: What's it about (purpose and general idea), what's a kernel
[x] SVM: Relation to classification
[x] ML: What does the regularization do (what is the intuition) (was already there!)
[x] SVM: Kernel intuition ("what is the kernel trick"), figure of gaussian and its scope
[x] Solution/PCA, first paragraph: "How. What is the generative model"
[x] Include results in report
[x] Minor things (spelling etc)
[x] Ref to PCA explanation (youtube video)
[ ] Miss-classification results earlier -> derive conclusion that the model is under informed.

Low priority

[ ] Variance retained: Also show non-accumulated plot
[ ] 2D projection: Try other dimensions as well
[ ] Dan put a question mark at the formula after "From linear algebra, we know that..."
[ ] ML intro: Not linear, general? Some function
[ ] What is positive semi-definite vectors? Relevans to positive length and variance of vectro.

AndersHqst / estimating-gaze-direction

Address feedback from Dan #28