Machine Learning With Arduino

[RubenSanchez]

Friends:

I am working on an implementation of the best know KNN (K- Nearest Neighbors). I am working with the MEGA2560.

The idea is develop a machine learning library using Arduino, in this case, for KNN algorithm.

Nowadays, the main functions are available, and it works! I can classify some data. I use the Iris Data Set from UCI.

If somebody is interested to join me and write the library, please be free to contact me. I would like keep this project alive!

Thank you very mucho, my email is sanchez...@gmail.com.

[Amulya Sahoo]

Hello Ruben, 

I want to join you in writing the library. Though I don't have knowledge of Machine Learning, I want to learn that and work on it. 

Regards

Amulya Kumar Sahoo
http://about.me/amulya349

--
You received this message because you are subscribed to the Google Groups "Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to developers+...@arduino.cc.

[Parth Chadha]

Hi Ruben,

I would like to contribute to the same. I have been working with Arduino/Raspberry Pi platforms for some time and have been recently attending Machine learning courses. I would love to work on developing a library for the same! 

Regards,

Parth Chadha

https://parthchadha.wordpress.com

[atajahangir moshayedi]

Hi 

i think its great idea .

is it really possible to have such this work like image processing with arduino?

--

You received this message because you are subscribed to the Google Groups "Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to developers+...@arduino.cc.

--

------------------------------------   

Sincerely yours
Dr.Ata Jahangir .Moshayedi
Ph.D Electronic and Robotic

    DOES_UOP_INDIA _(DEPARTMENT OF ELECTRONIC SCIENCE UNIVERSITY OF PUNE )

   MSC. Instrumentation
   BE. Power electrical

• IEEE _ Graduate Student Member 
• Life Time (LM) member of society of " INSTRUMENTATION  OF  INDIA"
• Annual member of Iranian association of electrical  and electronic engineers

--------------------
Post address: DEPARTMENT OF ELECTRONIC SCIENCE  

                          University of Pune, 

                           Pune Ganeshkhind, 
                          Pune-411007 ,
                           Maharashtra, (India) 
Phone nos :+91-20-25699841,+91-20-25691256,+91-20-25601419
E-mail& Web page: 

Academic:mosh...@electronics.unipune.ac.in ,

              moshayed...@ieee.org

Personal :  mosh...@gmail.com,  

" Be less curious about people and more curious about ideas " -- Marie Curie

 نیکو پدری گفت به فرزند خلف             دنیا طلبی به هند و عقبی به نجف

ور زانکه  نه دنیا و نه عقبی خواهی       بنشین به صفاهان که شود عمر تلف 

[Jose Ruben Sanchez Lopez]

Dear Dr. Ata:

Thank you very much for writing. Well, there are multiple algorithms for machine learning, some of them are expensives in memory and time. KNN is an algorithm that use a lot of memory in the microcontroller, but don't create a model of the data base, but we can use this for classify data.

My idea is program this algorithm in a microcontroller for has a plataform for classify objects (sensors play a importan rol) maybe using an arduino board like an agent only for classify data.

I think that we can use a similar approach for image processing in arduino. In my thesis work, I developed an algorithm for  image segmentation using the "KNN approach". Is simple and I think that could  works in arduino.

In the attached file, you can found the source code, sorry for write the functions in spanish, I will translate this.

In the attached you can see a .rar file, this rar has 2 folders. In knnarduino3 you can found a simple implementation of the algorithm, for get a class from data you must send "{0.1, 0.2, 0.3, 0.4}", this  is the format. You can see more data in iris.xlsx. The k parameter of the algorithm is 11, and you can change this for get better or worst classification results.

In the other hand, for knnarduino4, is a program that gets the best "k" parameter. This function computes the "recall" result for evaluating a "k". A recall near of 100% is desired. A k with best recall is returned in the serial interface. This part of the algorithm can be understood like "training-part" is a little ambicious but is only for naming.

Please be free for write me for any doubt.

Use the code for your projects and send me bugs and errors for improving.

Regards.

Atentamente

------------------------------------------------
  José Rubén Sánchez López
------------------------------------------------                                              

  sanchez...@gmail.com       

  Cel. 235 1002426                             

  SkypeID: ruben2048                      

------------------------------------------------

[Sancerio Limiardi]

Hi Ruben:

I am interested to know more about this project. Do you set this project up in GitHub or somewhere else? Thanks.