Re: enviroCar code challenge completed

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: enviroCar code challenge completed

prajwal m r
Hello,
This is Prajwal here. I understand its already late , but I wanted some more information regarding a few things before i could finalize my gsoc proposal.

I have labeled two sections A and B and numbered each question to make it easy for you to answer them.

I request you to please help my understand a few things here.

Section A :

The api "https://envirocar.org/api/stable/tracks" gives the following json which is an object containing an array "tracks" of 100 objects :
  
"tracks": [
  
{
"id": "58c24888268d1b08d804f99a",
"length": 12.892415419329199,
"sensor": 
{
"type": "car",
"properties": 
{
"engineDisplacement": 1598,
"model": "Astra",
"id": "574fc99de4b09078f97d721f",
"fuelType": "gasoline",
"constructionYear": 2001,
"manufacturer": "Opel"
}
}
},


.
.
.

]

Doubts : 

1)Here ,  does each object in the track array represent parameters for different users?
2)what exactly does the attribute length represent?


Section B

In the parameters shown in the following api:

which gives a json object of the form :
{
"type": "FeatureCollection",
"properties": 
{
"id": "58c24888268d1b08d804f99a",
"sensor": 
{
"type": "car",
"properties": 
{
"engineDisplacement": 1598,
"model": "Astra",
"id": "574fc99de4b09078f97d721f",
"fuelType": "gasoline",
"constructionYear": 2001,
"manufacturer": "Opel"
}
}
,
"length": 12.892415419329199
}
,
"features": [
  
{
"type": "Feature",
"geometry": 
{
...
}
,
"properties": 
{
...
}
}
,
  
{
...
}
,
  
{
...
}
,
  
{
...
}
,

Doubts:

1)What exactly does output of tracks/:trackid api endpoint mean? Is it the in detail data of a particular user on a given track?

2)In the coding challenge , I had to filter out the records based on the length parameter. now the question is even in the actual project will records be decided as outliers based on length only? or do we consider any other parameters too?

3)This has an object called "features". Does this represent the data of all the like Consumption ,Intake Pressure... etc
 at different instances of time in a journey? (as it contains coordinates for each record) 

I request you to please help me out with these questions.

Looking forward to hearing from you 

On Wed, Mar 22, 2017 at 8:54 PM, prajwal m r <[hidden email]> wrote:
Hello,
I would love to have two more clarifications(A and B) at this point.

A) So to summarize , can we say that the major tasks at hand are:
    1) Developing a ML model for outlier detection.
    2)Creation of a more lightweight GeoJSON output format that is supported by third party processing libraries ( As mentioned on the Ideas     Page ).
    Please do tell me if I have missed out any other major tasks that would be desirable to accomplish.

B) If I have understood correctly , Is the creation of a more lightweight GeoJSON output format necessary before going further with actual development of a ML model? In the sense , Would the ML model need to work on the new lighter output format? Please do correct me if I am wrong about this.

Looking forward for the inputs.

Thank you
-Prajwal 

On Wed, Mar 22, 2017 at 7:57 PM, Christoph Stasch <[hidden email]> wrote:
Dear Prajwal,

you could also use another programming language like R or python to implement the outlier filtering.

Best regards,

Christoph

2017-03-22 10:25 GMT+01:00 prajwal m r <[hidden email]>:
Hello,
I have a question regarding implementation . I can see that the server is written in Java. My doubt is should the machine learning code for filters for outlier detection also be in Java? Or could it also be written in other languages like python or R?

Thank you
-Prajwal

On Tue, Mar 14, 2017 at 4:51 PM, prajwal m r <[hidden email]> wrote:
Sure. Thank you so much.

On Tue, Mar 14, 2017 at 12:38 AM, Arne de Wall <[hidden email]> wrote:

Hello Prajwal,

we like your idea of applying machine learning on enviroCar data to detect some models and patterns in this type of sensor data. This was already an idea we had before but we never had the time to tackle this issue. It is a little bit beyond the scope of the original project description, but that is okay and we appreciate it ;) GSoC proposals are not limited to the description given by us.

However, your idea of training models and using it for alerting drivers when any specific "event" occurs is a little bit too ambitious and research intensive for the duration of GSoC. This type of driver assistance is a new field we have never tackled before and it contains a huge technology stack to understand and to implement. Therefore, we would give you the advise to go ahead with your machine learning idea and focus on a specific issue you like and think about the final outcome and the way you want to technically realize it. 

In your case it is not absolutely necessary to collect data by yourself, but you should get familiar with the overall architecture (especially data model + server) [0] in order to identify entry points for your specific solution. All data you need is freely available at [1]. Here you can access all the data of around 15000 tracks as open data. An example request looks like this [2]


Looking forward for your ideas.

Arne & Christoph & Matthes


[0] http://envirocar.github.io/enviroCar-server/

[1] https://envirocar.org/api/stable/

[2] https://envirocar.org/api/stable/tracks/58c24888268d1b08d804f99a

[3] https://github.com/enviroCar/envirocar-www-ng/tree/develop


Am 13.03.2017 um 10:26 schrieb prajwal m r:
Hello ,
Prajwal here. I had a few ideas for the gsoc project , " Quality Assurance in enviroCar". Please do tell me if these ideas are feasible to be implemented in the project :

1) We can develop a machine learning model to detect outliers. As the output results of enviroCar app would depend on various factors such  as experience of the driver (effective drivers have a smaller carbon footprint ) , the route chosen to travel (due to difference in traffic) , condition of the car (a well serviced car would probably  leave a smaller carbon footprint) , the type of car ( hatchback or an SUV ) ..etc , the ML model can be trained on all these factors and can be used to detect outliers on the go. Also we can keep training on all the correct data that keeps coming in. Hence it gets better at filtering data with time.

Additionally this machine learning model can also be used to alert the driver,  if any of the parameters of his car is out of normal range , and this can be personalized based on his/her car type , route taken ..etc ( Not directly related to quality assurance)

2) We can record all the parameters at the time when an erroneous output is generated (taking a snapshot of the parameters ). By doing this we can get to know if there is any correlation between any user actions and the erroneous results and hence this can be corrected.

Also , In the previous mail , You had mentioned that it is advisable to collect car data myself and experiment with it. But as i do not have a car , I request if i could get any previously collected real car data if possible to experiment with it. 

Looking forward for your feedback and guidance.

Thank you
-Prajwal

On Wed, Mar 8, 2017 at 2:28 PM, prajwal m r <[hidden email]> wrote:
Hello,
This is Prajwal here again. I found a few minor UI  bugs in the previous version of the enviroCar code challenge code that that I had submitted. Hence I have attached a new zip file containing debugged and refined code for the same. Please find the zip file attached with this mail.

Please rename "server.txt" to "server.js" after unzipping the folder.

This code can also be found on my git hub repository under this link : https://github.com/prajwalMR/enviroCar_code_challenge
  

Thank you
-Prajwal 


On Tue, Mar 7, 2017 at 7:51 PM, prajwal m r <[hidden email]> wrote:
Sure. Thanks a lot.

On Tue, Mar 7, 2017 at 3:40 PM, Arne de Wall <[hidden email]> wrote:

Dear Prajwal,

many thanks for your interest in the enviroCar Projekt and your submission of the corresponding coding challenge. We will review it and come back with feedback as soon as we have time and space for it.

Please be aware that the coding challenge is only one part of the GSoC application. The other part is the submission of a project proposal where you have to provide a clear workplan, clear milestones, and, most important, some innovative ideas for improving and ensuring the quality of the enviroCar data. The probably best way to achieve this, is to collect some data by yourself, get familiar with the enviroCar infrastructure and tools and to play around with the data.

Best regards,

Arne, Matthes & Christoph



Am 03.03.2017 um 12:56 schrieb prajwal m r:
Hello,

I am Prajwal , Doing my pre-final year of engineering in Computer Science in India.
 I have been working on the enviroCar code challenge from past couple of days, And I have completed it to the best of my knowledge.

Please find the zip file attached which contains all the relevant files of the code challenge.

***NOTE***
Please note that the server file has been RENAMED as " server.txt " In order to prevent gmail from blocking. This file has to to be renamed back to " server.js " of it to work.
************

Also , I would love to know how I could contribute to it further.

Looking forward to hearing from you. 

Thank you
-Prajwal










--

Dr. Christoph Stasch
52°North GmbH, Martin-Luther-King-Weg 24
48155 Muenster, Germany
tel. <a href="tel:%2B49%20%280%29251%20396371%20-32" value="+4925139637121" style="color:rgb(17,85,204);font-size:12.8px" target="_blank">+49 (0)251 396371 - 32
fax: <a href="tel:%2B49%20%280%29251%20396371%20-11" value="+4925139637111" style="color:rgb(17,85,204);font-size:12.8px" target="_blank">+49 (0)251 396371 -11
http://52north.org
twitter.@FiveTwoN



_______________________________________________
GSoC mailing list
[hidden email]
http://list.52north.org/mailman/listinfo/gsoc
Please respect our mailing list guidelines:
http://52north.org/resources/mailing-lists-and-forums/guidelines
Loading...