Category Archives: CARTO

Aggregating points: JSON on SQL and loops on infowindows

NOTE: I’ll use CARTO but you can apply all this to any webmapping technology backed by a modern database.

Get all the data

So we start with the typical use case where we have a one to many relationship like this:

    select e.cartodb_id,
           l.cartodb_id as locaction_id,
      from locations l
inner join employees e
        on e.location = l.location
  order by location

Easy peasy, we have a map with many stacked points. From here you can jump to this excellent post by James Milner about dense point maps. My example is not about having thousands of scattered points that at certain zoom levels overlap. Mine is a small set of locations but many points “stacking” on them. In this case you can do two things: aggregate or not. When you aggregate you pay a prize for readability: reducing all your data to those locations and maybe using visual variables to show counts or averages or any other aggregated value and finally try to use the interactivity of your map to complete the picture.

So at this point we have something like this map, no aggregation yet, but using transparency we can see where CARTO has many employees. We could also use a composite operation instead of transparency to modify the color of the stacked points.

Stacking points using transparency

Stacking points using transparency

Aggregate and count

OK, let’s do a GROUP BY the geometry and an aggregation like counting. At least now we know how many people are there but that’s all, we loose the rest of the details.

    select l.the_geom_webmercator,
           min(e.cartodb_id) as cartodb_id,
           count(1) as counts
      from locations l
inner join employees e
        on e.location = l.location
  group by l.the_geom_webmercator
Grouping by location and counting

Grouping by location and counting

Aggregate one field

But in my case, with CARTO we have PostgreSQL at hand so we can do way more than that. PostgreSQL has many many cool features, handling JSON types is one of them. Mix that with the fact that almost all template systems for front-end applications allow you to iterate over JavaScript Objects and you have a winner here.

So we can combine the json_agg function with MustacheJS iteration over objects to allow rendering the names of our employees.

    select l.the_geom_webmercator,
           min(e.cartodb_id) as cartodb_id,
           json_agg(e.firstname) as names, -- JSON aggregation
           count(1) as counts
      from locations l
inner join employees e
        on e.location = l.location
  group by l.the_geom_webmercator,l.location

And this bit of HTML and Mustache template to create a list of employees we can add to the infowindow template:

<ul style="margin:1em;list-style-type: disc;max-height:10em;">
{{#names}}<li class="CDB-infowindow-title">{{.}}</li>{{/names}}

List of employees on the infowindow

We could do this without JSON types, composing all the markup in the SQL statement but that’s generating quite a lot of content to move to the frontend and of course making the whole thing way harder to maintain.

Aggregate several fields

At this point we can repeat the same function for the rest of the fields but we need to iterate them separatedly. It’d be way better if we could create JSON objects with all the content we want to maintain in a single output field we could iterate on our infowindow. With PostgreSQL we can do this with the row_to_json function and nesting an inner double SELECT to give the properties names. We can use directly row_to_json(row(field1,field2,..)) but then our output fields would have generic names.

    select l.the_geom_webmercator,
           min(e.cartodb_id) as cartodb_id,
           count(1) as counts,
             SELECT r
               FROM (
                 SELECT photourl as photo,
                        coalesce(preferredname,firstname,'') as name
             ) r
           ),true)) as data
      from solutions.bamboo_locations l
inner join solutions.bamboo_employees e
        on e.location = l.location
  group by l.the_geom_webmercator,l.location
  order by counts asc

With this query now we have a data field with an array of objects with the display name and web address for the employee picture. Easy now to compose this in a simple infowindow where you can see the faces and names of my colleagues.

<div style="column-count:3;">
<span style="display:inline-block;margin-bottom:5px;">
  <img style="height:35px;" src="{{photo}}"/> 
  <span style="font-size:0.55em;">{{name}}</span>


Adding pictures and names

That’s it. You can do even more if you retrieve all the data directly from your database and render on the frontend, for example if you use D3 you probably can do fancy symbolizations and interactions.

One final note is that if you use UTF grids (like in these maps with CARTO) you need to be conservative with the amount of content you put on your interactivity because with medium and big datasets this can make your maps slow and too heavy for the front-end. On those cases you may want to change to an interactivity that works like WMS GetFeatureInfo workflow, where you retrieve the information directly from the backend when the user clicks on the map, instead of retrieving everything when loading your tiles.

Check the map below and how the interactions show the aggregated contents. What do you think of this technique? Any other procedure to display aggregated data that you think is more effective?

How a daily digest of geospatial links is distributed

TL;DR If you are interested on getting a daily digest of geospatial links subscribe to this mailing list or this atom feed. Take «daily» with a grain of salt.

Over the last six years Raf Roset, one of my favourite geonerds out there, has been sending all the cool stuff he founds about our geospatial world to Barcelona mailing list at OSGeo mailman server. He started circa 2011 sending one link per mail, but in 2013-04-03 he started to make a daily digest. A gun burst in Spanish is called Ráfaga so the joke was really at hand when someone proposed to call those digests that way.

Time passes, September 2014 and I ask Raf to send them also to Valencia mailing list, since most people there understand Catalan and the content was too good to be enjoyed only by our loved neighbours. Finally in January 2015 I decide to start translating them into Spanish and send them also to Spanish and Seville mailing lists.

Then in May I join CARTO and @jatorre thinks is a good idea if I can send them to the whole company mailing list so after some weeks I stop translating them into Spanish. Since that day I only do it English, trying to follow Raf lead everyday translating his mails and forwarding them to CARTO internal mailing list and the rest of the OSGeo ones.

Also at June I decided to put those mails in a simple website so the Ráfagas would also be accessible on GitHub and a static jekyll website so anyone could use the Atom feed to reach them.

Final chapter, in July I also decide to create a dedicated mailing list just for those people who are only interested in receiving those digest mails, obviously thinking in a broader audience, not just my fellow friends from Spain. I think at some point I will stop sending them to the Spanish lists because normally Ráfagas don’t fire any discussion and I’m sending the same message to three lists. To be fair they sometimes provoke discussions at CARTO mailing list. By the way I’m almost certain the full team has a filter to move them to their archives and they think I’m just an annoying spammer (a couple of times I’ve changed the subject just to troll them xDDD).

To conclude I want to post here my daily Ráfagas experience:

  • Raf is an early bird and sends the digest in the morning, I copy the contents into a shared Google Doc where a group of collaborators help me on translating the content. It may seem not a lot of effort, but doing this every single day needs a team. Really.
  • I go to my favorite text editor, put the translated content into a new file and start a local server to check the website renders properly.
  • If everything is OK I copy the rendered content and send it to CARTO and OSGeo mailing lists
  • I commit and Push to the GitHub repo so the website is updated along with the feed.
  • I archive Raf’s mail from my inbox.

Creating a Ráfaga

That’s it. Raf you are a formidable example of perseverance and I hope you’ll have the energy to keep giving us all those contents for many years. Thanks mate!

Creating a collaborative photo map: From Flickr to CARTO with Amazon Lambda

Phew, it’s been almost two years since my last techie blog post. I know I know, blame on me, I should’ve been writing more here but at least I did some nice posts at CARTO blog. Anyway, It’s Christmas today and because Internet is my playground and any piece of data I can put on a map can be a toy I spent a few hours having some geeky fun.

A few days ago a friend asked I knew any service to create a map of pictures in a collaborative environment. I thought maybe a trendy photo service like Instagram would be a good fit but it happens it’s super restrictive for developers so I headed to the good old Flickr. Flickr is one of those services that are really developer friendly, has a ton of super cool features and a decent mobile application and still, for some reason, it’s loosing its traction. Sad.

Anyway, Flickr has groups so a number of individuals can share geolocated pictures and they can display it on a map but sincerely, it has a very bad interface so probably we can do something better with CARTO. The issue then is how can we maintain an updated map in CARTO from a Flickr Group?

I’m a big fan of unmanaged services. I know there are people that love to maintain their servers but I’m not one of them. If I have to publish a website I try to use something static like uploading the site to Amazon S3 (i.e. my own website) or even better, use Github Pages like the Geoinquietos website. In this case not so long ago the only option to build an application to solve this issue was going to a PaaS service like Heroku, Amazon Beanstalk or Google App Engine, but they are meant for big applications typically involving a database and in general an architecture prepared for bigger things than this simple requirement. Over the last two years a new approach has emerged, a type of service that provides an automatically managed infrastructure to define small functions where each one is aimed to do a single functionality. They only live while they are being executed and afterwards the server is shut down. Amazon Lambda was the first of it’s class but recently also appeared Google Cloud Functions. On both services you can write your function in different languages (Python, Java, NodeJS, even PHP) and they can be triggered from a HTTP call or schedule its execution periodically.

As everything with Amazon, configuration from their website can be difficult and using it from the command line can be heroic. But it was a matter of time that something like Zappa would appear. Zappa is an application that makes deploying Python functions to Lambda dead easy. You basically configure a few settings and code your function and it takes care of the full cycle of deploying, versioning and even you can tail the logs from the cloud into your console in real time.

So to make this as short as possible, I coded a Lambda function that is exposed as a url acting as a proxy to Flickr API. This proxy will take URL arguments (or use some defaults) to retrieve photos information and will output them as a valid GeoJSON file. This allows me to create a CARTO Synchronized Table that updates every hour for example and retrieves the last pictures sent to a group (up to 500, as a Flickr API limitation). This dataset can then be used to create a BUILDER dashboard to present the pictures as nicely as possible.


Map for the last 500 pictures of the “Your Best Shot of 2016” group on Flickr

Additionally, on this map I kind of reverse geocoded image locations using a world borders dataset so I was able to add a country widget. Apart from that and a bit of CartoCSS to reproduce Flickr logo, the dashboard is quite simple. If you click on any of the images the pop up highlights the image (I love this feature) and you can go and visit the picture page in Flickr.


Pop up with the picture

But there are other methods on Flickr that returns photos, you can create a map of an account public uploads, or a map of the most interesting photos of the day, by photoset, etc. etc. All using the same proxy!!

I’ve created a github repo with the source code of this proxy (just around 130 lines of code) and more detailed instructions on how to set up your environment to deploy your own version of it on your account and use it in your own integrations. I have more ideas that I want to explore and I’ll try to share it here when I do them.

  • Leverage the Foursquare real time API to create a dataset in CARTO that is automatically updated every time I do a check-in
  • How to configure a scheduled task using the CARTO Engine SQL API. This is a super common use case when you develop projects with CARTO.
  • Create a CARTO Engine proxy to allow anonymous users to perform some tasks only accessible by default to authenticated users.

What do you think of this approach? Have you used Lambda for any other interesting use cases? Do you want me to continue posting on this topic about the other ideas I have? Feel free to comment here or reach me on twitter.