malariagen / ag1000g-phase3-data-paper

Other
1 stars 2 forks source link

Q57 - Phase 3 collection map figure #13

Closed cclarkson closed 4 years ago

cclarkson commented 4 years ago

Hi @hardingnj, @leehart, @alimanfoo

I've build three rough versions of the first panel (Western) to allow further discussion. Mali still needs some fine tuning, but you get the idea. "Western" is by far the most complex panel and if we can get something that works here, it will be fine for the other two.

Things to discuss here include:

review-notebook-app[bot] commented 4 years ago

Check out this pull request on  ReviewNB

Review Jupyter notebook visual diffs & provide feedback on notebooks.


Powered by ReviewNB

leehart commented 4 years ago

I think the country labels are OK actually. I would say The Gambia is the most ambiguous, but I don't think there's any real risk of a problem (confusion with Senegal or Mauritania).

I think using red circles for the sites might be a slight problem since red (of a different hue) is already being used to convey information about species.

Another visual problem with the site dots might be that it looks a bit like some of the pies are originating from those particular sites, or very close to, which might be confusing. I don't have a solution though.

I agree that the z-index of the site dots is an puzzle. It looks rough when they appear beneath the origin lines, but it will look odd if they appear above the origin lines or pies too.

I wonder if small green asterisks [Or maybe simple squares?] for the site locations would be better, to help distinguish them from the pies. It's not uncommon to see symbols on maps to indicate a type of site. (Don't use a cross!)

The scaled pies don't look too bad to me, and would counter the question "Why didn't you scale the pies to indicate sample size?".

I don't like the way the borders are thinner where the pies collide in Burkina Faso. I would also adjust the position of the Cote d'Ivoire pie so that it's clear that the origin line belongs to the pie, because it currently intersects with the land border.

The black borders and origin lines of the pies might stand out better if the black keyline of the map (land border) was slightly lighter, i.e. dark grey.

The text would probably look better if it had a white glow rather than a white box, but that might be asking too much! I'd be happy to look into that, but I suspect these map APIs are more difficult/restricted than the web-based APIs I've worked with.

It might look better if the pie's origin-lines taper to a point, like a pin or needle, rather than being just a simple line, but again that's just aesthetic polishing. It might make it clearer that it's pointing to a place on the map, rather than something else, like the country label. But it might not be worth the time investment.

cclarkson commented 4 years ago

Adds legends

hardingnj commented 4 years ago

Thanks Chris- looking great!

The black borders and origin lines of the pies might stand out better if the black keyline of the map (land border) was slightly lighter, i.e. dark grey.

Agree if possible

It might look better if the pie's origin-lines taper to a point, like a pin or needle, rather than being just a simple line, but again that's just aesthetic polishing. It might make it clearer that it's pointing to a place on the map, rather than something else, like the country label. But it might not be worth the time investment.

I think this is a nice idea... not sure how feasible it is though. A very narrow triangle maybe? Purely aesthetics though, would rather leave this for now.

cclarkson commented 4 years ago

Tweaks borders/outlines, <10 locations, adds full legend.

@hardingnj , @leehart - if you guys are happy with scaled pies (I think it looks cool), I will start the process of making it look neat again and apply to other regions?

Had a play with making the pie lines 'pointy', it doesn't look like there's a quick fix unfortunately. We can have 'arrows' but it looks like the overlapping points would actually be worse. As suggested, I could try plotting some kind of elongated triangles if folk think necessary?

alimanfoo commented 4 years ago

Looking good, agree with @hardingnj. Noticed legend should be "<10" rather than ">10".

Btw any way to make the <10 markers a little smaller/less prominent?

cclarkson commented 4 years ago

@alimanfoo - I've made the <10 symbols smaller (and fixed the legend - thanks!). They could be reduced further or have alpha lowered a bit more, what do you think of them now?

This is now a "finished" draft for the western panel, Mali is proving quite tricky aesthetically, so I'm going to come back to it after finishing central and eastern to try a few different layouts (and, of course, tweak the other western countries' pies a bit for good luck).

alimanfoo commented 4 years ago

@alimanfoo - I've made the <10 symbols smaller (and fixed the legend - thanks!). They could be reduced further or have alpha lowered a bit more, what do you think of them now?

What about using a crosshair symbol (+) for the <10? Might work better in Cameroon where there will be loads of points close by?

alimanfoo commented 4 years ago

Central map LGTM :+1:

Labelling of countries is difficult isn't it? I'm sure many readers would benefit from having country labels. But then I think if you have labels you have to put them within the country boundaries, which causes a problem for The Gambia. I don't have any clever ideas, just talking out loud.

hardingnj commented 4 years ago

Sorry- missed these updates earlier! Looking great I think. The 'x's work very well. I think the western map looks better without labels. It's hard to place The Gambia and G-B correctly. I think if we do it, the labels have to be directly over the centre and we fit everything else around them.

At the moment, weak preference to go without labels though.

leehart commented 4 years ago

When I said "Don't use a cross!" I was concerned about using conventional map symbols for churches. Buried pirate treasure is fine by me.

I still wonder if unfilled squares would be better, so I might mock those up to have a look.

I still don't see what all the fuss is about with the country labels. Do we really expect confusion or criticism there?

I'm still curious about using a white glow instead of a box on the labels, so I might try that out for us to have a look. I might see if I can get pointier pins while I'm at it. I might play with the line thickness on the circles and pins too.

Rounded rectangles for the colour swatches is interesting. Seems to work.

The spacing between the sample size circles in the legend and their labels is a bit inconsistent. I might try to fix that.

I'd be tempted to switch the order in the sample size legend too, so that the largest circle appears first, then the whitespace for the "Sample size" heading should look more consistent with the "Species" heading, or could be adjusted accordingly.

cclarkson commented 4 years ago

Hi @leehart, @hardingnj, @alimanfoo,

Here's a draft of all the map panels for comments/further work.

I'm a fan of the country labels but I have both versions of each here, I feel that people can work out what the labels are trying to convey. Plus due to the size of some countries the country labels don't even fit nicely without any pies/points. I've increased the pie surround line thickness to give it more pop (toning down the borders washed out the figure a bit).

I tried unfilled squares/circles but I felt they worked less well with overplotting and had to be slightly too large for the figure to look sensible. With shape markers we need to make clear which part points to the site, e.g. with the diamonds previously, it was the bottom point which signified the site.

@leehart, tweak away, I find the legend is a pain to space nicely so it would be cool if you find an easy way to make it look nicer.

leehart commented 4 years ago

For the time being draft PR #16 shows the 3 maps side by side (and one way to code that), currently deployed here https://malariagen.github.io/ag1000g-phase3-data-paper/PR/16 (as HTML) and here https://malariagen.github.io/ag1000g-phase3-data-paper/PR/16/manuscript.pdf (as PDF)

I branched off this PR's branch, but my draft PR is relative to master. (I couldn't see how to change the target branch, to indicate a more explicitly nested PR, i.e. a PR-off-a-PR.)

I'm still not sure of the best way to develop or integrate this, so @cclarkson should feel free to manually copy the code (or some version of that PoC) into this branch, however suits. I will probably test out & demo some map tweaks on my draft PR Thu/Fri, but that might amount to nothing. I understand that the planned destination for this figure is the 02.results.md section currently in PR #7

leehart commented 4 years ago

Oh... We didn't want to orientate this figure on a landscape page, did we? (That might prove tricky.)

alimanfoo commented 4 years ago

Oh... We didn't want to orientate this figure on a landscape page, did we? (That might prove tricky.)

No, I think we can't have landscape pages, so everything should be designed for portrait.

alimanfoo commented 4 years ago

You guys are gonna hate me, but... I think we've got room to just make this a single plot. I.e., just one panel, sub-saharan Africa.

leehart commented 4 years ago

OK, note to self: I'll try to combine these into one map, on this PR, tomorrow.

leehart commented 4 years ago
leehart commented 4 years ago

Todo:

Anything else?

alimanfoo commented 4 years ago

Looking very nice. Couple of very minor suggestions:

leehart commented 4 years ago

I've uploaded a new version, which should be deployed to the same location, with adjusted country name positions, which have the same font size as before - I haven't tried changing that yet. This version also includes a change that extends the map to the south, already, and I also cropped the east.

I need to figure out the maths for the sample size examples in the legend, which should be simple. I'm not sure what those circle sizes are currently based on. The pie radii are sqrt(nsamples * 0.01) but the legend currently uses markersize, e.g. markersize 25 = 150 samples.

leehart commented 4 years ago

Sample size legend now show true sizes, and n=10, n=50, n=100.

hardingnj commented 4 years ago

Looking good. Thanks :) I think perhaps we could be a little more consistent with how we orient the pointer lines, ie perpendicular to the coastline where possible. Angola and Gabon look a little off. Perhaps also a case for making the lines stand out a little relative to the borders.

Let's sign this off at the meeting today!

cclarkson commented 4 years ago

Looks great, I'm liking the white 'glow' around the country labels. Just a couple of comments:

leehart commented 4 years ago

Ha, I purposely made the pie outlines and arrows thinner, thinking it looked better, and I actually preferred the pie outlines removed completely, except that the pie's nice shadow effect did not seem to work well at large sizes. But I will have to suffer democracy, I suppose. I didn't personally have a problem with the black lines against the geographic borders since those were switched to gray rather than black. I'll screenshot a version without the pie borders to see if that's generally preferred.

leehart commented 4 years ago

I think perhaps we could be a little more consistent with how we orient the pointer lines, ie perpendicular to the coastline where possible. Angola and Gabon look a little off. Perhaps also a case for making the lines stand out a little relative to the borders.

I don't understand why the pointer line orientation needs to be consistent in that way. Gabon is pointing inland because there's lots of space available there, otherwise it clashes with Bioko. Similar for Angola, where the inland orientation keeps the label away from the legend and fills up gaping space. But I can appreciate those cases might look unusual.

Any feelings on that @cclarkson ?

leehart commented 4 years ago
  • maybe just a little tweaking of 'arrow' angles/pie positions (e.g. Angola, Cote d'Ivoire).

What is it about those that you would tweak?

leehart commented 4 years ago

Here's a screenshot of a version without the pie borders, for your consideration: image

leehart commented 4 years ago

I might tone down the legend labels too. They're looking a bit too bold in that shot.

cclarkson commented 4 years ago

If we are now down to a myriad opinions on small features, perhaps we have the perfect plot!

Angola, yep, same as Nick, feels like it should go out to sea. Cote d'Ivoire, yep, horizontal and vertical pointers feel wrong for some reason, less organic.

As for line weights, I think that really is a personal thing. Currently I feel the figure is "muted", there's no thick black lines drawing the eye anywhere. In fact my eye is first drawn to the legend with the larger bold font and larger colour blocks. That's not a criticism at all, it makes the figure look very cohesive and perhaps that is where the eye should go first. When I've been playing with these maps recently I seem to settle on more black around pies/arrows to make them pop out a bit, as I felt that is where I wanted the viewer to concentrate.

leehart commented 4 years ago

I didn't have time to sort out the font styling, with Seaborn's set_context('paper'), etc., but I gather we don't want to go much bigger anyway, because the figure is already crowded, and we can't go any smaller, due to style guidelines. You can see in the screenshot below that the country labels are about the same size as the text in the figure's label, but I expect we'll need to revisit this again later. In the meantime, I'll look around for a quicker/easier tool for this task!

Let me know if I've missed anything.

image

leehart commented 4 years ago
leehart commented 4 years ago

lw=2 instead of lw=1 on the legend circles doesn't make much difference, visually. I reckon these circles should now be exactly the same footprints as the pies. One potential improvement might be to not fill the legend circles, to mitigate any silhouette effect. image