Visualization embeddings

[jisungyoon]

This issue is related to Fig 3. visualization of embedding(maybe). I tried the figure which is colored by continent, but I think this figure does not give an enough information.

What's your thought about this figure? @yy @murrayds

[yy]

I’d annotate country names. Continent names can go as a legend with the color key. Can the map rotated in any way to match the usual map we see?

[jisungyoon]

I’d annotate country names. Continent names can go as a legend with the color key.

Okay, that's a good idea

Can the map rotated in any way to match the usual map we see?

I can't catch your meaning. Do you mean that make it similar to real geographical map? Rotating the axis?

[murrayds]

It looks so much better!

Perhaps we can use a more "striking" color for South America and Africa so that they are easier to locate? Similarly, since the U.S. is mostly contained with a single cluster, we can give them a more boring color (maybe the grey that is currently being used for South America?)

I can't catch your meaning. Do you mean that make it similar to real geographical map? Rotating the axis?

That's my understanding. There seems to be a somewhat natural "Europe" - "Asia" axis, so maybe orient the graph such that the Asia clusters are on one side, and the Europe clusters on the other?

[jisungyoon]

Like this way? I rotate the axis

[murrayds]

Like this way? I rotate the axis

This seems better to me! and I like the colors more too, they makes it easier to see the Spain/Portugal - South America connection

[yy]

I think if you mirror-flip wrt x-axis, you get a similar arrangement to the actual globe (viewed from north pole).

[jisungyoon]

Ok, Awesome! I will try.

[jisungyoon]

[jisungyoon]

I did a mirror flip and results is an upper figure I'm adding annotations on countries that have more than 100 institutes right now, how many countries do I need to annotate?

[yy]

As many as possible. As long as it's not too messy.

[jisungyoon]

Any comment?

[yy]

portugal and other south american countries? Québec? I think we can still put way more. Whatever we can provide stories.

[yy]

Also, again, set the dimension of the figure first. That gives you the minimum font size that you can use. We probably want to use that minimum size. For instance, if we want to have at least one full-width figure (in supp or something), then we can go wtih a lot of details. Feel free to use abbreviated names or even flags

[jisungyoon]

portugal and other south american countries? Québec? I think we can still put way more. Whatever we can provide stories.

In this embedding, Quebec is not closed to France. Below chunk of canada is Quebec

[yy]

What's that European country right below Canada?

[jisungyoon]

Also, again, set the dimension of the figure first. That gives you the minimum font size that you can use. We probably want to use that minimum size. For instance, if we want to have at least one full-width figure (in supp or something), then we can go wtih a lot of details. Feel free to use abbreviated names or even flags

We think this figure as similar size at labor flow network paper

[jisungyoon]

What's that European country right below Canada?

Iran, Turkey

[murrayds]

We can also relabel some of the countries to save space. For example, "United States" -> "U.S.A.", "United Kingdom" -> "U.K.", "Russian Federation" -> "Russia", and "Korea, Republic of" -> "S. Korea"

[jisungyoon]

We can also relabel some of the countries to save space. For example, "United States" -> "U.S.A.", "United Kingdom" -> "U.K.", "Russian Federation" -> "Russia", and "Korea, Republic of" -> "S. Korea"

Ok, I will try that.

[jisungyoon]

And, I am planning to draw the enlarged version of the upper figure (like labor flow figure). Here are the candidates.

1. Enlarge USA only, and colored by region, and also enlarge Massachusetts state only, colored by org types
2. Enlarge language-related part (Brazil, Spain, Chile, Colombia, Portugal)
3. Enlarge surroundings of Switzerland (France, Germany, Italy...)

I think we can pick 2 of them.

[yy]

1 & 2

[jisungyoon]

update figure.

[jisungyoon]

I tried two region categories for u.s.a. figures. which one ie better? @yy @murrayds Based on Cesus Based on economic

[jisungyoon]

please Ignore the color and positions of legend, I will fix it after determining standards for coloring. This is a figure that colored by states for reference

[murrayds]

Looking at the now, I think I like the Census colors (# 1) better, assuming that there will be state- or city-level labels for some key clusters. I like the more granular economic clusters (# 2), but they might not be as meaningful to non-US readers.

[jisungyoon]

Any comment? @yy @murrayds

[jisungyoon]

An interesting point about state-level embedding is the institution of D.C. are very broadly distributed. And, It seems likes to connect relatively weak area such as Hawaii or Alaska?

[murrayds]

Great! Is there a coherent cluster for Florida? It's a big population center so it can also be labeled.

An interesting point about state-level embedding is the institution of D.C. are very broadly distributed. And, It seems likes to connect relatively weak area such as Hawaii or Alaska?

Are they primarily government/military organizations? For example, It's likely that researchers for NASA will publish with the agency affiliation, which links to D.C., even if they work elsewhere in the country

[jisungyoon]

Are they primarily government/military organizations? For example, It's likely that researchers for NASA will publish with the agency affiliation, which links to D.C., even if they work elsewhere in the country

If we enlarge only D.C. entities, there are storing but small communities which are composed of universities (George Washington, Howard univ), and other inst looks like distributed broadly.

[jisungyoon]

Is there a coherent cluster for Florida? It's a big population center so it can also be labeled

Florida has two clusters, but the main cluster is located in the center. I will label Florida for the main cluster.

[jisungyoon]

I'm not familiar with states in the USA, can you make a list of states that I didn't include on the upper figure, but you are interested in? @murrayds

[jisungyoon]

I added labels as many as possible. Any comment?

[jisungyoon]

It is almost a top 22 states based on population

[yy]

try to avoid using the lines. There is no one-to-one correspondence between the state and the vectors anyway. try to put the label where the points are located. A typo for Florida btw. What're the points at the bottom?

[jisungyoon]

try to avoid using the lines. There is no one-to-one correspondence between the state and the vectors anyway. try to put the label where the points are located. A typo for Florida btw.

Is it okay to decrease the font size? I tried to put the label where the points are located, but it is a little bit big for a given space.

What're the points at the bottom?

Another Flordia. The University of Miami, and hospital, health research inst of The University of Miami,

[yy]

It depends on the figure size. https://www.nature.com/documents/nature-final-artwork.pdf nature's minimum font size is actually pretty small (5pt), although we probably want to keep larger than that.

I'd put Florida twice. (also the leftmost group?)

[jisungyoon]

you mean leftmost of Northeast group? That's university in Pittsburgh, ex) CMU, Univ of Pittsburgh,

institutions of Pennsylvaniy is divided into two clear clusters, non-Pittsburgh (Most of them are located in Philadelphia) and Pittsburgh

[jisungyoon]

This is a test plot for Massachusetts, Most of the institutes which are located on the lower-right part are Institutes in Cambridge, Boston. Ant upper-left parts are institutes in Worcester or public school of Massachusetts.

[murrayds]

I think that, with Florida, most of the major states in terms of university systems, are labeled.

The zoom-in on Boston looks pretty good, we can also compare it to New York state to see which is most interesting.

Its current hard to tell what is an isntitute and what is a univeristy. Maybe make the colors for institute and university more distinct? Or make the points bigger?

[jisungyoon]

The zoom-in on Boston looks pretty good, we can also compare it to New York state to see which is most interesting.

Do you mean in the next papers?

Its current hard to tell what is an isntitute and what is a univeristy. Maybe make the colors for institute and university more distinct? Or make the points bigger?

Yeah, I will re-arrange the color, and also resize the points

[jisungyoon]

Update

[murrayds]

The zoom-in on Boston looks pretty good, we can also compare it to New York state to see which is most interesting.

Do you mean in the next papers?

Yeah, after thinking about it, Boston is a natural choice for this paper. We can potentially examine others in future papers.

I like the new figure! What does the size of points mean? And maybe label one or tow more hospitals?

[jisungyoon]

What does the size of points mean?

It is inst_size but logged.

And maybe label one or tow more hospitals?

Do you know the famous hospitals in Massachusetts? and also research institutes

[murrayds]

Ah, that makes sense.

Perhaps "Massachusetts General Hospital" and "Boston Medical Center"?

It would also be good to label 'Tufts University" and "Brandeis University"—both well-known schools but not entirely in the same sphere as MIT and Harvard

[jisungyoon]

There is too many Bostons, so I added Umass health care

[murrayds]

There is too many Bostons, so I added Umass health care

Lol, fair enough, I think its looking pretty good!

[jisungyoon]

Any comment?

[murrayds]

Looks good! Just a couple of spelling fixes:

"Flordia" -> "Florida" "Virgina" -> "Virginia" "Coloardo" -> "Colorado" "New Maxico" -> "New Mexico"

[jisungyoon]

Looks good! Just a couple of spelling fixes:

"Flordia" -> "Florida" "Virgina" -> "Virginia" "Coloardo" -> "Colorado" "New Maxico" -> "New Mexico"

Sorry, I will fix it.

[jisungyoon]

And, another thing is I used window-size=2, embedding_dim=200 for this visualization Do we need to change the embedding? Because I saw you changed the dim with 128, 256 recently @murrayds