From Visualization to Analysis
Formal Metadata
Title 
From Visualization to Analysis

Subtitle 
Stop using heatmaps to discover spatial patterns

Title of Series  
Author 

Contributors 

License 
CC Attribution 3.0 Germany:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor. 
Identifiers 

Publisher 

Release Date 
2019

Language 
English

Content Metadata
Subject Area  
Abstract 
It is a common thing when pattern analysis cases are just about visualizing the features. The talk would start with a gentle introduction into pattern analysis algorithms. I would focus on the benefits of using analysis in comparison to visualizing methods, present cases and show pitfalls in usage and implementing these algorithms.

Keywords  General 
Related Material
Video is cited by the following resource
00:00
Presentation of a group
Multiplication sign
Moment (mathematics)
Mathematical analysis
Coma Berenices
Product (business)
2 (number)
Product (business)
Data management
Causality
Visualization (computer graphics)
Energy level
Spectrum (functional analysis)
01:20
Point (geometry)
Area
Presentation of a group
Statistics
Touchscreen
Mapping
Distribution (mathematics)
Point (geometry)
Moment (mathematics)
Mathematical analysis
Mathematical analysis
Mereology
Uniform resource locator
Population density
Software
Visualization (computer graphics)
Audiovisualisierung
Point cloud
Cuboid
Pattern language
Data structure
Matrix (mathematics)
Resultant
Computing platform
03:17
Point (geometry)
Threedimensional space
Digital filter
Mobile app
Distribution (mathematics)
Decision theory
Cellular automaton
Set (mathematics)
Mathematical analysis
Mereology
Number
Usability
Local Group
Force
Vector space
Visualization (computer graphics)
Square number
Audiovisualisierung
Absolute value
Task (computing)
Social class
Area
Distribution (mathematics)
Dialect
Mapping
Point (geometry)
Mathematical analysis
Tablet computer
Type theory
Arithmetic mean
Vector space
Universe (mathematics)
Pattern language
Boundary value problem
Mathematical optimization
Resultant
Reading (process)
Audiovisualisierung
05:49
Area
Cluster sampling
Dialect
Mapping
Key (cryptography)
Decision theory
Forcing (mathematics)
Mathematical analysis
Time series
Sound effect
Bit
Bit rate
Mathematical analysis
Line (geometry)
Mereology
Graph coloring
Number
Local Group
Different (Kate Ryan album)
Pattern language
Office suite
Resultant
Social class
Chisquared distribution
08:16
Email
Group action
Building
Open source
Multiplication sign
Execution unit
Range (statistics)
Sheaf (mathematics)
Set (mathematics)
Time series
Data analysis
Distance
Mereology
Number
Latent heat
Population density
Delay differential equation
Pattern language
Representation (politics)
Spacetime
Extension (kinesiology)
Social class
Task (computing)
Area
Graph (mathematics)
Touchscreen
Mapping
Key (cryptography)
Temporal logic
Graph (mathematics)
Mathematical analysis
Range (statistics)
Line (geometry)
Funktionalanalysis
Term (mathematics)
Subject indexing
Latent heat
Population density
Arithmetic mean
Uniform resource locator
Uniformer Raum
Personal digital assistant
Function (mathematics)
Convex hull
Quicksort
Library (computing)
11:43
Point (geometry)
Presentation of a group
Point (geometry)
Uniform convergence
Distance
Area
Number
Type theory
Population density
Causality
Type theory
Circle
Uniformer Raum
Kernel (computing)
Quadratic equation
Cycle (graph theory)
Species
Resultant
Triangle
Data type
12:51
Cluster sampling
Dialect
Random number
Distribution (mathematics)
Decision theory
Outlier
Gene cluster
Set (mathematics)
Water vapor
Mathematical analysis
Area
Chain
Arithmetic mean
Set (mathematics)
Pattern language
Representation (politics)
Information
Process (computing)
Task (computing)
Social class
Distribution (mathematics)
Link (knot theory)
Decision theory
Outlier
Web page
Mathematical analysis
Statistics
Distance
Inclusion map
Type theory
Category of being
Vector space
Interface (computing)
Video game
Resultant
14:33
Revision control
Area
Chain
Dialect
Distribution (mathematics)
Decision theory
Personal digital assistant
Gradient
Moment (mathematics)
System call
Graph coloring
Sequence
15:48
Point (geometry)
Metre
Slide rule
Pixel
Real number
Set (mathematics)
Distance
Area
Number
Centralizer and normalizer
Term (mathematics)
Different (Kate Ryan album)
Wellformed formula
Radius
Cuboid
Pixel
MaÃŸ <Mathematik>
Physical system
Default (computer science)
Area
Distribution (mathematics)
Metre
Mapping
Projective plane
Counting
Personal digital assistant
Uniformer Raum
Point cloud
Cycle (graph theory)
18:09
Torus
Distribution (mathematics)
Outlier
Execution unit
Range (statistics)
Data analysis
Area
Mathematics
Different (Kate Ryan album)
Local ring
Social class
Graphics processing unit
Area
Algorithm
Mapping
Outlier
Software developer
Interior (topology)
Parameter (computer programming)
Measurement
Repository (publishing)
Website
Cycle (graph theory)
Point (geometry)
Surface
Slide rule
Trail
Open source
Link (knot theory)
Mathematical analysis
BEEP
Distance
Axonometric projection
Graph coloring
Product (business)
Network topology
Wellformed formula
Term (mathematics)
Energy level
Computing platform
Scaling (geometry)
Information
Military base
Mathematical analysis
Cartesian coordinate system
Software
Circle
Personal digital assistant
Calculation
Local ring
Distortion (mathematics)
Library (computing)
23:30
Cluster sampling
Implementation
Distribution (mathematics)
Outlier
MiniDisc
Parameter (computer programming)
Mathematical analysis
Local ring
Area
Dean number
Number
24:23
Cluster sampling
Presentation of a group
Distribution (mathematics)
Outlier
Parameter (computer programming)
Mathematical analysis
Client (computing)
Area
Twitter
Goodness of fit
Personal digital assistant
Hill differential equation
Moving average
Local ring
Audiovisualisierung
Dean number
25:54
Area
Cluster sampling
Presentation of a group
Scaling (geometry)
Kepler conjecture
Key (cryptography)
Mapping
Distribution (mathematics)
Outlier
Debugger
Parameter (computer programming)
Mathematical analysis
Client (computing)
Area
Mathematics
Personal digital assistant
Representation (politics)
Right angle
Local ring
Computing platform
Dean number
Audiovisualisierung
00:05
the engravings the e.u. to to fill their time for the next sixty seconds while we checked levels was around doing so i want to remind you of this is the most tightly packed and you just stole my seat and look at this article.
00:24
this is the most tightly packed session i see this entire conference i want to remind you really think back to would vastly said this morning smile of the person next to it cost you nothing to be kind to one another and easy with the elbows. so so i just want to have to take a moment to thank you all for coming and i give you a nickel a cause or they can and cannot. well hello everyone. that's a huge thanks a lot for coming here and i was not expecting to see sold my people here. now even the presentation is not very good at least they know how to create great topics so you know i hold me to work and gay so i'm product manager in a spectrum dot com i'm not sure if you know what it is but probably you a check than and try and what we are developing and yet actually what would be.
01:21
i'm talking about is the heat maps so just to clarify the heat maps that i would be talking about these on the right part of the screen so its frozen mall more like on your presentation of heat map for cartographers but there is like common statistical methods to visualize future is a reason. left part so i would be talking about the the is the right part so i hope nobody just decide to leave his moment so let's continue to work there is so the pipe is actually the year data visualization for easy density usually for for points density on the at on the map and. and here is a nice a like picture from somebody america phone can and blog on my box jail. and actually what's the reason to talk about is the heat maps town as i told we are developing a software that's the cloud based platform to work with location data and analysis and sold by analysis actually mean analysis we do provide some just a typical methods to work with its location data which allows to like. to discover a day to more than just visualizing it and we had a lot of requests concerning adding the heat map as of the result is a show method and i was wondering what's the reason for his ad because i actually do not like a lot of them but as i found a lot of businesses a lot of like our users and us. summers are actually using it quite a lot so we decided to check our expertise they are using them and to understand whether we need to implement this feature in our software and a here is a couple of pieces were when our customers wanted to use the heat map to discover some patterns and to find some.
03:08
a glass or is it said trust so i would just stop on not couple of them at here is the first one that's for joe marketing in our customer wanted to find the area was like the most attractive area within the city to spread their businesses in this area us they get some points of interest.
03:28
the points of attractions it was like somebody stops some schools so universities it said trade center and to they wanted to legalize it and actually they didn't act on the heat map here you can see and select part and my question question was surreal ease it what you wanted to see i mean there is nothing. you see here so you cannot make any decisions based on these huge map they'll for what we actually he made with them together we build a just a simple greet over is easy to area ad then counted the number of features for every square inch enron's of what analysis so the quarter to analysis allow us. this to understand not just the areas where is there is a high intense of futures but also to understand which distribution type is actually within this app like data said because there is no need to find class or is in the data set of the data said is not lost early distributed so therefore. they were satisfied enough was a solution and did these bowman to i decided to go deeper into other cases and we've got to the second one here is the visualization of one hundred thousand points on the map and did we also the task was to define the areas where people that came from.
04:49
into the peninsular so as you can see on the heat rather a lot of late read regions here and you cannot actually compare themselves they look pretty much the same as is a lot of various but what we've done we actually also build the vector agreed and counted the number of futures per every said. there is any used in a bog g.l. rendering to also have to go to these rectory into the third dimension therefore it was easier to compare is the absolute values begin with and features and you could see that the slate read by it's actually quite different from from missouri far. it's so comparing these two result is a she knew me see that the heat map is really simplifies of the usual eyes and features and the truly do not show the patterns eighty six pectin to be shown by using as a result is a shunt techniques and here we used invested means gasification which is quite good.
05:50
for the long tail he's to graham and they'll for it even the collar sweet used is like to fuel the color ramp also show even better is and the huge maps which is so used to the rail ball. four m. for each i would be talking about to be played our side is also light at the pain i believe for every car to go for it he said. the third at peace we also met in ukraine there is a car accidents map and we each way used to say to define solution to find decisions where to prove that the true can patrol strong candidates the to read or or.
06:37
for police officers which are actually staying on the road in trying to like to find was there is a car is moving with the in the speed and the mets and as you could see his like the whole cities in a mess so i have no idea how the just try to. to find the the places to put disease had better also and here is our map was ample lines at which shows were actually as he put the spectral sea and on the our map we also have made a great but instead of just counting the number of features of iran's the. the classes and golf lesson alice is also known as local morons i analysis though for this analysis allows you to see the class herself hywel use last are so full well use as well as cauliflowers so if there are some or all that parts will each have. a bit different number of accidents compare into the nearest one it would be the childlike there and as you can see comparing these two results actually the main patterns remain in the same so the middle part of the city's read as well as in the heat map but using the analysis techniques you could see use it. there are some specific regions where it is significantly higher number of accidents and it is recommended to put petrol sexually and this area so you see hear and see in also part of the cd and also here and the south part of the cd. that was also the keys and the force kesa that we also mad during his is he to map so requests and was a bollard the i was a bozzi time series analysis so that's also about the effect art director data we were founded the issue was too.
08:29
to find a class or he saw for crimes analysis and to realize it on the map is into heat map by means they wanted to find the sparks of crimes in some specific areas but want to be of diana and what we are also talking about that or keys and data analysis is not just about maps that it's also about charts the toilets. also about some in texas and here you could see chart does it show and actually usable line is the likely his to her i'm number of features are some specific time range and the red line is that the number or the nearest neighbour in the uk so we chose at to what extent the. futures are cured are close to each other so it shows the number of gloucester to these features therefore is you can see at the high well year of our this case the heat a big number of thefts and as are crimes does not mean that they actually taking part in one n one terry to. very and eighty's really hard to find some classes just by visualizing the future is using the heat map and then run and eighteen to z. he's to graham at therefore when the euro got to the earth's nearest neighbor index you could see you was theirs areas like some awesome. but that the graphs go down and it means that some features are these being and it is buoyant some crimes locations tended to be closer to each other and it is of the next step to create a map is always asian and actually year at find these cluster but it's it's not a task so is small. and what you're talking about is as there is no need to start with the map but you should start with slick as or analysis techniques and then moved to the map just to find the place and here i would share is the library's it we developed for building in the year at times or his analysis that's more easier to a deal. the analysis on the daytime range so you could just it's like time free to break so there is no need to define your own times sections it's much easier to work with your entire time at a time series and then divided into the year alan sector so feel free to go there and war. the key to its open source but definitely at and it would be as an instant last flight also at a gay so what actually is there is enough not using heat maps yet. a good start and what about the map was how it works so here is a representation of the history and for some data sets so you could see these are some features here and later the black lines and how the heat map works it actually build some sort of the carol function is usually is a galaxy on fire. action and then it all to place and create these density a graph reaches actually represent in on the screen off the map. even here you can see that even there are some features are actual located quiet in a quite a distance the it does not work very well and here you would see answer next light that's the beautiful also if using the heat map at its units not probably the better is the best light of but say it shows how it could work wrong and.
11:54
just imagine if they're us just like the number of features or is there is one feature or for each of the cycle you would see for some women for some specific radio so the influence they would be like to place when the you would have the highest value of four is the scary will function in the territory.
12:14
the leash is not actually has any futures so that's probably the main thing that you should think about and would not work with the like on the local at all and even these species are reserved for a presentation as a uniform and colonel but for any type of carrying all you would find his distance. these were under local at all your ted's is overlap and misleading so your research would not shows the appropriate results and it's this point you'd better be not to use of visualizing it also because its it would harm your your results at four is that i would recommend of the first by and if there is a task to.
12:55
some klosters to runs in your neighbor analysis just to understand whether at the distribution types cluster because he was there is a tough to find clusters we need to understand it and that the data set his class surgeons and the next is an ex actually task is to find a way whereas the squatters so when you're running nearest. the analysis i you're getting the results and then you could drown think what are its analysis and findings a distribution type at it is the vector agreed of course it also has some problems for as a year like working because the what are its size its steel and as a question i would also state. stop on the task to buy a to at least there and all are over life and features and to these issue could be fixed by using waters analysis techniques and.
13:48
another one if you're not just working for the only research is intense of data about your also want to find the extra the class or is and i'll fly heiress within your data set i am recommending to use the classes and outliers analysis. each would returns the data into four categories is actually the clusters hywel use classrooms low low well yes and then i'll fly arse of why are some in the well is surrounded by all yours or low well we're surrounded by hywel years and within these so the whole data said french you could normalized all the features. and find the veterans and one all the data set and continue working with your representation and a decision based on the analysis and he would be much more sufficient to to to move forward.
14:40
yet here are some cases the this this is a vivid the next one at just this crazy.
14:49
that's the color. i would not stop a lot of this moment but the general idea just do not to use the rail ball because it's really not a very good idea i am attach invading to the great article by exertion lot to last about the issue based gradient of the main idea is that it is really not very good for people. more people try to identify these. they made all regions between the main main collar see the red green and and blue so it would not work very well for people's i to understand these call our ramp and therefore it is not very good of course for people with certain color blindness therefore it is a much more he commanded. to use one hugh or even to hughes just too late to show all these day version and or a sequential at data distribution and yet area in france there is another pain because at first the also said the e.u. we're a cloud base.
15:57
titian so we are most folks and on other cloud based solutions and for a lot of cases they are using pixels and as a kind of us the systems to render these heat map and is you may imagine pixel is nothing in jet tearing injure graphical term so you could not to just because souls actually two. to any territory to understand reach areas a neighbour in and they could and should influence and influence on each other and other hand it's quite easy to more from like big disaster to some small local issues that could be you know just fixed by anyone so it is very much. many people a to have to work with the heat maps and he doesn't since it does not have any geographical background as the distances as a neighbour in distances he teases very often misused by journeys by some propagandist cetera et cetera so that's the reason it is not the best idea. use it another been about pixels also and where the.
17:05
the neighbour in distance so pixels as a previous slide is the settings that he's like showings area of the influence but. we each area of interest should be using actually there's a question and in most cases you are just working with some kind of beauty just imagine so ok let's take five hundred meters but why ass i was there is a common usage through command the distance to understand what is the distance between. in points should be considered as neighboring distance and the distance of the influence therefore is there is a formula that is from nearby but to a it takes into account a number of features so this data central count and also the area of the ball going to have distribution so you if you're working with the data set on the c.t.a. you. we should not take the bond in a box because you know the the area would be huge difference and also a you should to just find surreal area of the territory. another one sorry he's at the met projection is also bars the pixel somehow these point to you could see you for example here is a huge cycle or is a europe and you may imagines it actually does not work is a real wall world like these because we should count distances on the earth's surface.
18:28
as though for its should be a huge distortion of this cycle it should not be a cycle because for sas products that people usually use the market or so the market for us year as therefore he does not work very well indeed also misleads the researcher because on disease global scale a your future is built it would overlap. but on surreal the census would be completed the friend asked so here is the slide for his the overlap in not overlap in but change is a distortion change in the cirrus therefore and what we use in our analysis because we are building the platform for like everyone i know we're using been sent to form a laugh or. the ad. measuring distances which works quite well on our website and where we actually open sourced of these formula and based algorithms to measure distances between nearest points so we could be an easy implemented in any solution and it to use the like nearest neighbor analysis based.
19:31
on the calculations based on cuba so it's on the g.p.u. and it works quite well and i am recommending to read the article for my friend and also be at the back and developer of the product a soul what he released is actually the old glories terms that works extremely. really fast it's building the poultry and and works and going downtown to find the nearest neighbor and the distances nearest neighbor and it works extremely fast so you could count the distances to like three solve three million futures within one minute on the at and if you guess far is and always was ten sixty. yet so it's not the newest one as so here he said the the beep repository and also the top so feel free to use it in any applications and yeah that's the general information to sum up what i have but i was talking about so first of all before you. use in the the visualize and methods to find class terrorists you should understand the wiser izzy's data set these class really distributed therefore i recommend into use in your stable analysis you can find and you could use these are ever represent tory for his even sent to cuba and any also use in appropriate to. area of intense spare amateur and measuring distance at and that's the formula that was also all its lights because gap area of interest in france should be a geographical unit for abusers agent also do not to use a rail ball use the wine or to hugh color ramp for visualizing it's much easier. to understand the differences between ass a middle middle middle range is also a use the local more inside an analysis to define classes and outliers eat may a normalised all the data and he may show some local changes between the in the middle of the inside the data says. which shows the insides of the data also. i did not mention in this lies but if you have some way to heat map so you are using some column in your and by its data said to build the heat map it's better to use the get his aura to july star analysis just to find this area swiss hywel use and low well as it works very well and is it is also could be used. i use in to buy sell its the library for easy data analysis here is an extremely i believe valuable links for as the interest in the stories research she said terry said try and do you may see the open source stuff there to the euro countries. should get and then to lot of hope it was interesting somehow somebody yeah. you are. will be what cases would you consider appropriate for using heat map sexually. i like a and yet i believe he would argue with good work we have a lot of data on local level i mean if you're walking working with the indoor data map in it could work quite well because you've got a lot of. like to keep it on not g.p.'s are indoor tracks and so there is not like overlap and you show and there is no link distortion is this year as because of the local also probably that would work as. so what. if additional feature agreeable the day it must stuff just some comments above their the use allies are looking more and for example especially for the network why use agreed approach when he studied in net why i think it's possible to apply directly that a lot more.
23:41
morons. and for bees so yeah. thanks very good question. um yeah i think it would be better if we can to number of features breach road segment number of my future is in discs these accidents yeah.
24:04
or. yeah probably this was a reason because those the current to implementation so yeah that's definitely that should be done so thanks. it all. but yeah i would repeat the question of it.
24:28
it is. a few years. the age of your compensation china good for our economic trends continue to the island. thanks the just i would try to understand by myself to your question was about the presentation and all the mats for it.
24:59
but it so. so yeah i know that he was or was he doing. he said i did that for the baby and the reason those who are. first the case is probably as in most yeah so the question was about the how differently presented different to those asian approaches it could use instead of him absent how our clients and our are actually. take it into consideration so they are i'm not sure but yeah for the we are yet just from from behind we're going to see solution but frozen be to be cases the two year using its steel commons that they.
25:54
we are trying to. our clients use our the specialists to to build the keys to get their souls they're quite tricky when we are talking to them that it's not probably the best solution for you let's go to another sedition but unfortunately it's just for a couple of cases because we could not communicate with every. one definitely so. leave right and some articles and making presentations what are you would a lot to do more from there. another. about the heat not been agreed the representation when you can scale with the heat map people like like it i think because the changes according to the scale which i i hate but it did and did don't like the great because you just see one big. he said when your area a high skills and perhaps that's a how the seat. i think you're actually i think it could be fixed and it's already fixed in camp larger all it is it will brussels asian platform so they do is the segregation i mean the greed is those asian on the front on the front and therefore it could be also set up to the. like on fly a career ender in the greek but we still in the kepler geology are using the soda markets are so we are facing the same issue with the year was a great size but it's interesting to created on the front end i think and to run it directly and the client side it would be on the flight. so is technically it's possible i think it would be war he thanks a lot to thank you very much over and over and of applause room we thank you.