1. The Data - Service Requests

20 Jun 2019

[A Data Science Doodle]

Introduction

This post is the first in a series, The Data Science Doodle, chronicling my journey deliberately, informally getting into Data Science. I plan to write experiences, learnings and report on projects I work on. At the writing of this post, I am working through the R for Data Science book and ‘concurrently’ exploring a dataset of interest. The idea is to evaluate my thought process and workflows, see how, what I learn can be applied to my real world.

The Project

The City of Cape Town makes some of its data publicly available on its Open Data Portal. There’s a plethora of data themes from which one can draw insights about that city from a public service perspective. My focus is on a dataset of Service Requests. Several questions and conclusions can be drawn by digging deeper into this data. I am to understand this data better and see how R can be used to process and manipulate it.

A Brief on Service Requests

The City of Cape Town’s C3 Notification system (now 2019, referred to as Service Requests) was introduced in 2007 to enable the municipality to better manage and resolve residents’ complaints. The success of the system was acknowledged when the City received an Impact Award for Innovation in the Public Sector at the Africa SAP User Group (AFSUG) in 2011.

It is an internal process that is used to record, track and report complaints and requests from residents and ratepayers. There are about 900 different complaint types ranging from potholes, water leaks, power outages and muggings, to employee pay queries or internal maintenance requests.

When residents contact the City, a notification is created on the C3 system. All possible types of complaints for the various different City Departments are catalogued (dated, categorised and geo-coded). The complainant is then given a reference number, which allows them to follow up on the complaint. The notification will be closed as soon as the complaint has been dealt with.

The City’s Call Centre can be contacted by using several channels; calls, email, sms, city’s website, facebook, twitter, App - Transport for Cape Town.

References 1, 2, 3, 4.

Digging-In

The Service requests data can be downloaded in Microsoft Excel format or ODS. To investigate the data I resorted to LibreCalc and OpenRefine ~ a free, open source, powerful tool for working with messy data.

Alex Petralia puts it out well

OpenRefine is designed for messy data. Said differently, if you have clean data that simply needs to be reorganized, you’re better off using Microsoft Excel, R, SAS, Python pandas or virtually any other database software.

Data Explore

OpenRefine(OR) can work with .ods files but, I struggled to load the January 2011 file. I even increased OpenRefine’s default memory (from 1024 to 2048) setting, still there was struggle loading the file. Opening the file in LibreCalc revealed that the first four rows in the month’s services request records were descriptive fields which even included merged cells. I deleted these and exported the file to text (csv) format.

After the export the file loaded like a breeze in OpenRefine.

I repeated the .ods file cleaning procedure in LibreCalc for all the remaining files February to December 2011. I then imported all 12 csv files into OpenRefine. Again memory was an issue and I increase the value to 4096.

Cleaning Up

All fields where imported as as string/ text into OR. I then renamed and redefined data types of the field names to have some structure.

Notes:

• (*) planning to index the suburb field.
• x_coord, y_coord: set to text for now as focus on them will be much later.
• notification_date: Is truly date, formatting will be later.
• notification_id: is the unique request identifier. Assigned type (BIG)INT (so as to avoid problems later when the data grows - Hint from the web, Digital Ocean.)

Aside: Now, is it work_centre or work_center ? well, it’s command center!

The following operations where applied to the fields to clean it somewhat in OR;

• Trimmed white spaces.
• Collapsed consecutive white spaces.
• Convert all suburb names to uppercase.
• Transformed notification_id To Number
• Reformatted notification_date to the format YYYY.MM.DD .(The plan being to later import the data into a PostgreSQL database.)

Some OR transformation expressions and functions are documented here.

For the date transformations I took a hint from here and using the expression

 value.slice(6, 10) + '.' + value.slice(3, 5) + '.' + value.slice(0, 2)

another GREL expression which was widely applied to populate a field with values from another field

cells["work_centre"].value

Messy Suburbs

The most time consuming stage was cleaning the names of the suburbs. Why suburb? As a “geo” person, the mind is wired to think geocoding and suburbs is a good reference. The (x_coord, y_coord) pair for this dataset was largely unassigned which led to the focus on suburb.

The Cluster and Edit Operation was mostly used. The variations in the names was very wide. Hinting to a lack of standardisation on the names and most likely alluding to the use of ‘free-text’ in the initial capture of the Service Requests.

Some records had suburb value ‘UNASSIGNED’. For suburbs which did not ‘make sense’ in a local context, like London, UK, I assigned INVALID.

Taking Stock (A Database)

As an intermediate stage I exported the records as SQL from OR to a sqlite database. cct_service_requests.sqlite3

[Aside: “use .sqlite3 since that is most descriptive of what version of SQLite is needed to work with the database”;Tools: DB Browser for SQLite Version 3.10.1 (Qt Verson 5.7.1, SQL Cipher Version 3.15.2)]

Exporting the records to SQL from OR resulted in Out of Memory issues even with 6GB dedicated to OR and the data with 906501 records. I resorted to importing a CSV instead.

I was looking at retaining suburbs with atleast 5 or more records or less if there was a corresponding suburb defined. Those with any less were assigned to the larger area boundary. To get insight into the suburbs I used the SQL Query

SELECT count(suburb) as a, suburb
 FROM service_requests_2011
    GROUP BY suburb
    ORDER BY a DESC;

There were 3659 unique suburbs. Largely with a count of one, of which on inspection were clearly a result of typos during data entry.

Having done the identification of non-unique suburbs. The task was now to identify these in OR and clean them up.

Getting Help

Somewhere along the processing I came across ‘noise’ in the data which gave me memory issues in OR and spreadsheet programs. This had caused the data file to balloon to 800MB+.

I resorted to twitter for help

In summary the technique to clean the data was

I used Facet -> Customized facets -> Text length Facet on the ‘suburb’ column, then adjusted the filter to remove the highest value which was a single huge outlier…

The problem record entry was notification_id = 1003477951. I deleted this and proceeded with the cleaning.

Readying

After further cleaning there were now 1100 unique suburbs with the highest count of 4.

The Next Stage was to match this service request data with the spatial reference frame.

At this stage one can start doing miscellaneous analysis of the 2011 service requests. Distribution per month, per suburb, most requested service, etc.

#PostScript

True to the “80% is spend in data cleaning” assertion, this portion of the project took a lot of time! Next up is the preparation of spatial data for the suburbs. Using the suburbs list from service_requests_2011, collate a corresponding spatial dataset.

Button Pushing GIS Analyst, Not

07 May 2019

SCT Part 5

Loading…

When the year began, I resolved to blog at least twice a quarter. That has proved to be too ambitious. The ‘distraction’ has largely been a dirty dataset I was, still am, cleaning. You see, I am now a wannabe Geo-Data Scientist. This hype and perception of self has been influenced by endless periods of watching Pyception. I am proudly, progressively going through the book, R for Data Science, affectionately known as R4DS. Honestly one of the spurring factors are the cool and sophisticated looking graphs made with ggplot.

Map Monkey

Since learning about it, I have tried as much as possible to not myself mold into a ‘Map Monkey’ or ‘Button-Pushing’ GIS Analyst. This year I have taken up Data Science, the buzzword of the moment right? I see M.L. and A.I. are trending actually. As a geospatial specialist there is demand for one to be a ‘data manager’ already. Map production or results from a spatial analysis demand one massage data in one way or the other. Taking on stuff about data and some stats shouldn’t be too foreign. So in the first quarter of 2019 I was setting up my machine for Data Science and studying the same. It was pleasant to discover that Geographic Data Science is actually a thing and I am not being very divergent from years I have already invested in work and study.

So here’s a peek of what’s on my PC.

I chose Chocolatey to spare me hours of troubleshooting dependency, local installation paths issues with Node, Yarn, et. al. I must mention though that having DBeaver running wasn’t straight forward. The solution came from here and some of the screamed at screen shots are shown below.

I cannot, not mention kepler.gl, the reason I have Yarn up there. Kepler.gl is any spatial data visualisation enthusiast’s goldmine. You look like a pro with very little effort. I have gone through various blog posts and video tutorials to try and Think Like (a) Git with minimum success. So hopefully my learning R accompanied by version control with the help of GitHub for Desktop should help. See, I have been abusing GitHub for just an online storage and blog hosting space…none of the version control functionality.

Endearing Geo

I still love my geo, so when I hit a wall with the Data Science ‘things’, I fire up QGIS and dabble my spatial data in PostGIS. As one who is a tinkerer, I picked somewhere that SQL skills are an important must have for a geo person, moreso Spatial SQL. ( This article makes a great read on the importance of SQL for a geo person and how to get started. My blog chronicles my personal path of the same).

EnterpriseDB distros of PostgreSQL really make the installation and configuration of PostGIS a breeze. Within minutes one has access to a functional sandbox and production ready space. Through click click, type type I had PostGIS ready.

DBeaver has become a dear companion navigating the SQL land, with the plus of a progressively improving spatial view. DB Manager in QGIS is a great go-to when QGIS if fired up.

Still I cannot just dash over the ability to view spatial from within a database (DBeaver) environment. PgAdmin 4 has a spatial viewer but a little more Googling tends to put one off when it comes to the design of PgAdmin 4. There also have been efforts to have a simplistic spatial viewer, my favourite being PostGIS Preview, although I haven’t been able to make it work yet. Sometimes it is good to be able to see ‘map’ data without having to fire up a GIS program. Which makes spatial data view in DBeaver celebratable …enforcing the notion of “spatial is just another table in the database” aka spatial is not special. Well, it is special when you need to know about EPSG codes and why, when to use which.

New and shiny get’s my attention, so I have also tried Azure Data Studio just for the preview PostgreSQL extension. This release was apparently a big thing . I experimented with it a bit, connected to my PostGIS and decided to settle on DBeaver. “Be really good at one thing”, right?

#PostScript

So I’m off to learning R, SQL and git just to be Button Pushing GIS-Analyst, Not. The next blog post should be about Data Science, Geo Data Science.

A Bright Square Kilometre

31 Dec 2018

SCT Part 4

All That Glitters

Image credit - Gathua’s Blog

The above is a typical representation of Cape Town night lighting. How would that look like from space? One way to answer that is demonstrated here but, the data used there is quite coarse. This post is on a similar approach with the further investigation of finding the brightest spot on a Cape Town night. Data on the location of street lights was used as provided on the Cape Town open data portal.

Some Assumptions

• Street lighting is the only lighting at night time at a place.
• All street lights are represented (no broken light).
• The dataset is complete and accurate (temporally).
• All the lights have identical lumens.
• [Other implied assumptions.]

Public Lighting

The subject data is described as “Location of street light poles and area lighting (high-masts) in the Cape Town metropolitan area.” - Public Lighting 2017.zip

In Q, I ran Statistics for text field to find out how many unique lighting types there were. Just two - Streetlight and High-mast Light. A total of 229 538 lights.

The distinctiveness of street light types persuaded me to give greater weight to the High-mast Light. So, an additional assumption -

High-mast Lighting is twice as bright as just Streetlight lighting.

So in Q I created a weight field and populated it appropriately.

To scratch the curiosity itch. The lighting in Q, for the CBD and surrounds.

Boot Strapped Visualisation

If you haven’t heard of Kepler.gl you must check it out. I handles your geodata in the browser and does some amazing things.

First step is to convert the shapefile to csv, json or geojson. In Q this is achieved via right click on the layer and Save As.

Load the CapeTown_Public_Lighting.csv in Kepler.gl

(On loading the CapeTown_Public_Lighting.csv in Kepler.gl the weight field was being interpreted as type Boolean and not just Integer. It turns out this was a bug (December 2018). I quickly remembered this as I am actively ‘watching’ the Kepler.gl github repository)

As a workaround I used weights of 100 and 200 for the StreetLights and High-mast Light respectively.

Panning around the map, the above caught my eye. Turns out the more isolated dots are high-mast lights. Actually in Kepler, you can filter/ view a type at a time or together using different colours and a whole other options.

Our objective of finding the brightest spot requires we aggregate the points data. That is home turf for Kepler. With a few clicks we have this;

We are looking for The Brightest Square Kilometre and we factor in the assumption high-mast light = 2 X Streetlight.

So in order to get to calculating the correct area for the hexagon being employed by Kepler, I got to asking.

Taking a hint from here and using the formula

 A = ½• n • r² • sin( (2π) / n)

where:

• A is the area of a polygon
• r is the outer radius
• n is the number of sides

I got the radius as 0.6204 km

Under Radius Tab, Enter Radius value of 0.6204 and decrease the Coverage (base area of the hexagon being visualised) Switching ON The Enable Height Tab - Enter a reasonably high Elevation Scale, The HeighT is based on Point Count. High Precision Rendering turned ON.

These parameters aid quick identification of the ‘tallest’ hexagonal column hence the brightest square kilometre. With a count of 923.

Light Weight

To take into account the weight of the light type (high-mast vs street light), we add another data layer (a duplicate of the CapeTown_Public_Lighting.csv really), apply a weight filter Add a Layer based on this and style it with similar parameters to the first Layer. (At the time of writing this post, a filter when applied to a dataset affects all the layers based on it. Hence the need to duplicate the dataset with this approach.)

With a highest count of 52 per the square kilometre. Even doubling high mast lighting for impact falls short of the 871 (923 - 52) of ‘StreetLight’ alone.

Now styling the High Mast Lighting with grey scale…so we can see the two together.

There is apparent correlation - Highly lit areas also have high mast lights. High Mast ~ ‘white’ columns.

You can interact with the Street Lights Map below.

#PostScript - Replicate

Kepler.gl is under active development and new features are being added continuously including some fascinating ones on the developers’ road map.

One of the features is the ability to share a visualisation easily. So here goes:

• The Data and Settings ~ for this exercise can be downloaded here.

• Download the file to your local drive and open it from The Kepler.gl Website.

Cape Flood Safe

27 Dec 2018

SCT Part 3

(18.42281, -33.95947)

So, those are the numbers (coordinates) to keep and punch into the satellite navigator in the unlikely event of the Atlantic Ocean bursting its Cape Town shores. If you stay in Cape Town, the first place that comes to mind when you hear ‘Flood Survival’ is likely Table Mountain and rightly so. But, which spot exactly can one stand longest on solid earth in the event of a massive flood?

Cut To The Chase: The Safe Strip

How To DEM

Get Data

Relief maps always get my attention. The mystery of attempting to model reality on the computer enthuses me. There are plenty tutorials on the net on how to do this so here’s my version.

I got some vector base data (Metro Boundary, Suburbs, Roads, Railway Line, Building Footprints and a DEM - Digital Elevation Model ) off City of Cape Town’s Open Data Portal or here. The vector data, I loaded into a geopackage. How? Of special interest was the DEM, described as “Digital model (10m Grid) depicting the elevation of the geographical surface (Bare Earth Model) of the Cape Town municipal area.” The Open Data Portal has the data stored as 10m_Grid_GeoTiff.zip.

Style Terrain

After extracting the compressed elevation data, I loaded it in QGIS.

1. Load 10m_BA,
2. Style the ‘relief map’ in Properties –> Style

The raster would then look as shown below

1. Now to create a hillshade. Do Raster –> Analysis –>DEM (Terrain Models)…

   With the settings shown below create a Hillshade DEM

This gives us a hillshade …

1. To get us a Terrain Map. Set transparency in the Relief Map (10M_BA). RightClick –> Properties –> Transparency

1. Now a combined view of The Relief and Hillshade …

1. Now Load the other support vector layers
   • Metro Boundary (Flood Plane)
   • Suburbs
   • Roads
   • Railway Line
   • Building Footprints

Getting 3D

To get started with 3D Terrain. Load and activate the Qgis2threejs Plugin.(Read more). (This is one approach to getting 3D in QGIS. At the time of writing this post, 3D comes native with QGIS. Read (more.). But, I was using QGIS 2.18.7, Portable gotten from here.

The Qgis2threejs Plugin is unique in that it bootstraps the process to have a web ready 3D model to play with.

1. Launching the Plugin for the first time should give something like this …

1. Zoom to Cape Town CBD and include all of the table of Table Mountain.

Now, to prepare for exporting the subject area, emulate the following series of settings paying attention to the parts World, DEM, Roads,

The Surburbs Layer is used mainly for labelling purposes in the final export. We are not interested in showing the Suburbs boundaries for now.

Through Trial and Error and guided by the absolute height of the DEM, we get the optimum height value, 1 060m, to use for our Flood Plane (Which in essence is just a polygon covering all of our area of interest - Metro Boundary in this case).

Export To Web

In order to have an interactive 3D Map simulating a flood.

1. Ensure 3DViewer(dat-gui).html is chosen under the Template file in the Qgis2threejs Plugin.

2. Choose the appropriate path and name for the index html for the visualisation. In this case CCTerrain.html is chosen.

3. Export and Qgis2threejs will generate the necessary style sheets and other files for the export.

4. Now using a text editor, edit parts of the exported html file (CTTerrain.html) to reflect what the project is about and put in some usage information.

A Preview below - full screen here

#PostScript - Find The Safe Spot

Click around the map to turn on/off the layers.

Relations, I Mean, Tables Galore

03 Jul 2018

SCT Part 2

Embracing Text

In the previous blog, we explored how we can load (explicitly) spatial data, initially in shapefile format, to a geopackage. In this post we explore the loading of relations a.k.a table to the geopackage. Again, data from the Open Data Portal is used.

Data Is Rarely Clean!

We will explore “Data showing City land that has been disposed of.” The data is spread over several spreadsheets covering various time periods. In some instance yearly quarters. The data custodian is given as “Manager: Property Development & Acquisitions city of capetown.”

In the data wrangling process using LibreOfficeCalc, the open source equivalent of MS Excel. Several attributes will be learnt about the data. Especially that which hinders a refined data structure. Issue will be varied, among them:

• Not all records being filled in.
• No consistency to period of registrations.
• Inconsistent formatting of currency.
• Sale price included or excluded VAT inconsistently.
• Typos in area names.
• A regions field which definition could not be got from the Open Data Portal.

To improve the data we may end up

• Adding some fields to improve schema
• Concluding area field = Suburb Name (per suburb_layer from previous blog post)
• Matching the area field names to match up with Suburb Name in the Base Suburbs reference data we adopted.This procedure inadvertently modifies data from it’s original structure. However, knowledge of local suburb boundaries and place locations means the impact will be minimised.

Casual inspection reveals

• A lot of sales to churches and trusts. Possible follow-up questions would be. Was this influenced by any legislation in a particular year? Which year saw the largest number of sales to churches?

Loading Tabular Data To Geopackage

With the Land Disposals data, covering several years, curated and consolidated into one spreadsheet. It is time to import this data into our geopackage. Here are the steps to do it:

1. From the Spreadsheet application (OpenOfficeCalc), Save file as and choose Text CSV (Comma Separated Values).

To load the Text File to the geopackage we use any one of the two approaches.

Method 1: Using QGIS

1. Start Q if not already running.

2. Select Add Delimited Text. This is a large comma icon with a (+)plus sign. When you load the file, it will likely look like the screenshot below. (Note that the image shown here is of a file that would have undergone considerable pre-processing. Especially on field names and data types in cells.)

Make the selections as shown above. In particular ‘No Geometry’.

Opening the table in Q we see the data we just imported.

Now let’s save this to the GeoPackage.

3. Rick Click the Layer name, –> Save As. Then specify the location of the geopackage. In the dialog box that comes up, under Geometry, choose No Geometry. (We are just importing a non-spatial table.).

That’s it!

Let’s explore an alternative way of getting the text file into the GeoPackage.

Method 2: Using SQLite Browser

1. Open the GeoPackage in the SQLite Database Browser.

2. Do File –> Import –> Table from CSV. Check on; Column Names in first line and Trim fields?. Separator is comma.

Done!

3 Switch to the Browse Data tab to see what we have just loaded.

A Review: What To Choose

On inspecting the table loaded with QGIS and SQLite Browser. Somehow Q correctly and auto-magically distinguish text and numeric field data types. SQLite Browser made everything type “TEXT” in this particular case.

Let’s experiments some more with our data.

Close SQlite Browser and switch to QGIS. (It is good practice to close one application when accessing data from another. Particularly if you are to do editing.)

Launch DB Manager and Browse to the geopackage. Let’s see how many distinct suburbs do we have in this table. Start the SQL Windows and run…

SELECT DISTINCT  area FROM  land_disposals_2009_2018
ORDER by area ASC;

we get 102 records . We see there is NULL in first entry so we really have 101 records. On close inspection we notice there are a lot of typos in some names.

We will need to clean this data if we are to have meaningful analysis later on.

Let’s check the suburbs layer and see how the data compares

SELECT DISTINCT  NAME FROM  suburbs
ORDER by NAME ASC;

We get 773! This is gonna be interesting, especially getting exact matches.

Before cleaning the data further, lets see if we can ‘connect’ the suburbs and land_disposals datasets.

SELECT  * FROM suburbs
JOIN land_disposals_2009_2018
ON suburbs.NAME = land_disposals_2009_2018.area;

Yep! We have some hits.

We are going to be basing our analysis on the suburbs layer so we use that layer for a base and format subsequent datasets after that it as much as possible. Cleaning can be done in QGIS user interface editing or better still in the sqlite database (geopackage) using SQL.

#PostScript

The take away is that GeoPackage can store tables, no sweat.

The join operation used above is simplistic. One would need to appropriately use LEFT, RIGHT, INNER JOIN.

I used:

• QGIS 2.18 (DB Manager)
• DB Browser for SQLite 3.10.1
• SublimeText 3.1.1