Beau ------------------

SECTION 1 --------------------

Overall this way a very clear and well written section. I think
there's a few specifics in terms of notation and linking back to
previous information that lost me when I was reading, but these should
be easily fixed from the following section. It's a very clear
explanation of the different folds assignments, although it could help
to make more specific links between them.


SECTION 2 --------------------


Sec 2: Paragraph 1: Introduce Cross Validation, then V-fold CV, then
explain the components effecting the method.

	Sentence two of the first paragraph: In terms of the
	topic/stress position concept we learned, it may make sense to
	switch the order here: "With the traning set, we estimate
	model parameters and use the validation set to assess model
	performance", or something like this.

	Could probably split sentence 3 into two: The first explaining
	V-fold, and the second how it works with networks. Right now
	it appears to be both explaining V-fold AND how this matches
	up in the network situation.

Sec 2: Paragraph 2: Explain the lay-out of the paper.

	No comments, this is a very smooth paragraph.
	
Sec 1: Paragraph 3: Set up the notation

	I was a little bit confused in sentence 2. What exactly is Y?
	Is the matrix Y, or is Y a list of observed edges?

	In the pseudocode, we have Models M_1 ... M_t. This is (I
	think) the first we have heard of these models. Are these
	models in section 1? Similarly, this is the first we've heard
	of the parameters and later variables.

	It seems that for each model, we split the data into v-folds,
	compute the error on each and average them. But I'm not
	positive quite yet this is the set-up

Sec 2.1, Paragraph 1: 
	
	Nice pictures in figure 2. It may help to have labels
	underneath the images with "edgeCV", "latincv", etc.... since
	this was the label presented in the paragraph. As a reader, I
	went to figure two with the "edgeCV" label in mind.


Sec 2.2, Paragraph 1:
	
	The natural progression here is that the simplest model,
	uniform assignment to folds, has issues in that it may assign
	every edge containing node i to a single fold. It may help to
	contrast this here, especially to link section 2.1 with the
	current section. "In contrast to edgeCV, latinCV guarantees
	each row and column has an equal number of occurences..."

Sec 2.2, Paragraph 2: 

	This is a tricky concept to explain, but I think references
	the images REALLY helps here. When the reader get's lost
	trying to keep all of the details in their head, they can
	reference the image to get a grasp of what is happening.

Sec 2.3, Paragraph 1:

	Since this is the first time we've introduced L, it makes
	sense to put this in the stress position! At first, I didn't
issue	quite understand that each NODE goes to a fold, since the
	previous two sections either dealt with blocks or individual
	cells.

	"thus the training set sizes are comparable to those in V 2
	fold random edge CV" I stumbled a bit on this sentence. The
	main idea is that the training size are about the same as
	those described in sections 2.1 and 2.2, it might be more
	clear to the reader to reference this specifically.

	In the final sentence, it may help to add a sentence
	explaining WHY this is the most important consequence of the
	assignment feature. In this, we understand that it's
	important, but as a reader I'm not sure why this relationship
	should be so striking.