Re: [Sugar-devel] GSoC Proposal ML activity
https://drive.google.com/open?id=1dSnO-2E2SmaV__RPu7ZqQ7TqzH92rfMI I added that part on this document with more details, please do read. I am sorry for the mess I keep forgetting to Reply All. Anyway I would like to submit the proposal, does it seem good or is there any changes needed or parts added? On Sun, 31 Mar 2019 at 23:31, Walter Bender wrote: > > > On Fri, Mar 29, 2019 at 2:31 PM Ahmed ElSabbagh < > ahmed.h.elsabb...@gmail.com> wrote: > >> Hi Walter, >> I Added an illustration of how I imagine it would look like. >> I don't exactly understand your question regarding how Algorithm work. >> > > I am asking, how are you going to apply tensor flow to solve the problem > you are describing? > > >> As for CPU intensity, I have to ask how low-end are we talking about? I >> am have not used tensorflow before, but this google experiment (from which >> I draw inspiration for this project) >> https://experiments.withgoogle.com/teachable-machine, needs only less >> than 100 images to differentiate between 3 classes, and it runs fast on >> browser, it should in theory be able train the model on relatively slow >> computer (although with a bit difficulty). >> https://drive.google.com/open?id=1UJUclV_0otspq0KIYK_ms4wUkdA4VHBc >> >> >> On Fri, 29 Mar 2019 at 15:29, Walter Bender >> wrote: >> >>> Interesting idea. Could be a fun activity. >>> A couple of comments: >>> (1) it is a bit thin on details of how you'd implement the app itself. >>> What would the interface look like? what sorts of controls would there be? >>> (2) Also, in brief, how does your algorithm work? How CPU intensive is >>> it? Realistic to run on low-end laptops? >>> >>> There was some work done at RIT about 5 years ago on a sign-language >>> chat for Sugar -- never completed. Might be worth investigating. If I can, >>> I will find you a link. >>> >>> regards. >>> >>> -walter >>> >>> On Fri, Mar 29, 2019 at 9:12 AM Ahmed ElSabbagh < >>> ahmed.h.elsabb...@gmail.com> wrote: >>> >>>> Hello I am sending a proposal for Sugar Activity. >>>> Can you please provide an honest opinion, do you think we can work on >>>> that or is my resume too underwhelming? >>>> Thank you in advance >>>> >>>> >>>> https://drive.google.com/file/d/1bCLeTv6fpfD71ExFwMtbB8WRExMspieg/view?usp=sharing >>>> >>>> ___ >>>> Sugar-devel mailing list >>>> Sugar-devel@lists.sugarlabs.org >>>> http://lists.sugarlabs.org/listinfo/sugar-devel >>>> >>> >>> >>> -- >>> Walter Bender >>> Sugar Labs >>> http://www.sugarlabs.org >>> <http://www.sugarlabs.org> >>> >> > > -- > Walter Bender > Sugar Labs > http://www.sugarlabs.org > <http://www.sugarlabs.org> > ___ Sugar-devel mailing list Sugar-devel@lists.sugarlabs.org http://lists.sugarlabs.org/listinfo/sugar-devel
[Sugar-devel] Fwd: GSoC Proposal ML activity
-- Forwarded message - From: Ahmed ElSabbagh Date: Wed, 3 Apr 2019 at 13:53 Subject: Re: [Sugar-devel] GSoC Proposal ML activity To: Walter Bender https://drive.google.com/open?id=1dSnO-2E2SmaV__RPu7ZqQ7TqzH92rfMI I added that part on this document with more details, please do read. On Sun, 31 Mar 2019 at 23:31, Walter Bender wrote: > > > On Fri, Mar 29, 2019 at 2:31 PM Ahmed ElSabbagh < > ahmed.h.elsabb...@gmail.com> wrote: > >> Hi Walter, >> I Added an illustration of how I imagine it would look like. >> I don't exactly understand your question regarding how Algorithm work. >> > > I am asking, how are you going to apply tensor flow to solve the problem > you are describing? > > >> As for CPU intensity, I have to ask how low-end are we talking about? I >> am have not used tensorflow before, but this google experiment (from which >> I draw inspiration for this project) >> https://experiments.withgoogle.com/teachable-machine, needs only less >> than 100 images to differentiate between 3 classes, and it runs fast on >> browser, it should in theory be able train the model on relatively slow >> computer (although with a bit difficulty). >> https://drive.google.com/open?id=1UJUclV_0otspq0KIYK_ms4wUkdA4VHBc >> >> >> On Fri, 29 Mar 2019 at 15:29, Walter Bender >> wrote: >> >>> Interesting idea. Could be a fun activity. >>> A couple of comments: >>> (1) it is a bit thin on details of how you'd implement the app itself. >>> What would the interface look like? what sorts of controls would there be? >>> (2) Also, in brief, how does your algorithm work? How CPU intensive is >>> it? Realistic to run on low-end laptops? >>> >>> There was some work done at RIT about 5 years ago on a sign-language >>> chat for Sugar -- never completed. Might be worth investigating. If I can, >>> I will find you a link. >>> >>> regards. >>> >>> -walter >>> >>> On Fri, Mar 29, 2019 at 9:12 AM Ahmed ElSabbagh < >>> ahmed.h.elsabb...@gmail.com> wrote: >>> >>>> Hello I am sending a proposal for Sugar Activity. >>>> Can you please provide an honest opinion, do you think we can work on >>>> that or is my resume too underwhelming? >>>> Thank you in advance >>>> >>>> >>>> https://drive.google.com/file/d/1bCLeTv6fpfD71ExFwMtbB8WRExMspieg/view?usp=sharing >>>> >>>> ___ >>>> Sugar-devel mailing list >>>> Sugar-devel@lists.sugarlabs.org >>>> http://lists.sugarlabs.org/listinfo/sugar-devel >>>> >>> >>> >>> -- >>> Walter Bender >>> Sugar Labs >>> http://www.sugarlabs.org >>> <http://www.sugarlabs.org> >>> >> > > -- > Walter Bender > Sugar Labs > http://www.sugarlabs.org > <http://www.sugarlabs.org> > ___ Sugar-devel mailing list Sugar-devel@lists.sugarlabs.org http://lists.sugarlabs.org/listinfo/sugar-devel
Re: [Sugar-devel] GSoC Proposal ML activity
Hi Walter, I Added an illustration of how I imagine it would look like. I don't exactly understand your question regarding how Algorithm work. As for CPU intensity, I have to ask how low-end are we talking about? I am have not used tensorflow before, but this google experiment (from which I draw inspiration for this project) https://experiments.withgoogle.com/teachable-machine, needs only less than 100 images to differentiate between 3 classes, and it runs fast on browser, it should in theory be able train the model on relatively slow computer (although with a bit difficulty). https://drive.google.com/open?id=1UJUclV_0otspq0KIYK_ms4wUkdA4VHBc On Fri, 29 Mar 2019 at 15:29, Walter Bender wrote: > Interesting idea. Could be a fun activity. > A couple of comments: > (1) it is a bit thin on details of how you'd implement the app itself. > What would the interface look like? what sorts of controls would there be? > (2) Also, in brief, how does your algorithm work? How CPU intensive is it? > Realistic to run on low-end laptops? > > There was some work done at RIT about 5 years ago on a sign-language chat > for Sugar -- never completed. Might be worth investigating. If I can, I > will find you a link. > > regards. > > -walter > > On Fri, Mar 29, 2019 at 9:12 AM Ahmed ElSabbagh < > ahmed.h.elsabb...@gmail.com> wrote: > >> Hello I am sending a proposal for Sugar Activity. >> Can you please provide an honest opinion, do you think we can work on >> that or is my resume too underwhelming? >> Thank you in advance >> >> >> https://drive.google.com/file/d/1bCLeTv6fpfD71ExFwMtbB8WRExMspieg/view?usp=sharing >> >> ___ >> Sugar-devel mailing list >> Sugar-devel@lists.sugarlabs.org >> http://lists.sugarlabs.org/listinfo/sugar-devel >> > > > -- > Walter Bender > Sugar Labs > http://www.sugarlabs.org > <http://www.sugarlabs.org> > ___ Sugar-devel mailing list Sugar-devel@lists.sugarlabs.org http://lists.sugarlabs.org/listinfo/sugar-devel
[Sugar-devel] GSoC Proposal ML activity
Hello I am sending a proposal for Sugar Activity. Can you please provide an honest opinion, do you think we can work on that or is my resume too underwhelming? Thank you in advance https://drive.google.com/file/d/1bCLeTv6fpfD71ExFwMtbB8WRExMspieg/view?usp=sharing ___ Sugar-devel mailing list Sugar-devel@lists.sugarlabs.org http://lists.sugarlabs.org/listinfo/sugar-devel
Re: [Sugar-devel] ML idea discussion
I must admit James that I have don't know for sure how performance could be measured and I have been trying to find proper way to measure it. One thing I know for sure however is that measurement units are supposed to be standardized and rarely change https://en.wikipedia.org/wiki/Computer_performance. The problem either way would to turn whatever performance unit given into a unit of time given the size of the data set before starting the training process, and this is my point of research at the moment. On Thu, 28 Mar 2019 at 10:01, James Cameron wrote: > Thanks. Are there standard measures of hardware performance you could > use as estimates? Measurement units change over time. At the moment > I'm most familiar with Geekbench multi-core processor scores. > > On Thu, Mar 28, 2019 at 09:09:53AM +0200, Ahmed ElSabbagh wrote: > > As for hardware, I am not sure how strong should it be to run a training > model > > for Neural Network or any other classifier. > > > > As for internet access, the training can be done offline no problem, it > can be > > slow depending on the size of the data set they add, it could be a > feature to > > try predict or estimate the amount of time it will take them given a > minimum > > hardware specification, the problem however will be the lack of pictures > to > > use, so a minimum number of photos and categories can be provided > offline at > > any given time, with other pictures and categories available downloadable > > whenever internet is available. > > > > On Thu, 28 Mar 2019 at 02:41, James Cameron <[1]qu...@laptop.org> wrote: > > > > Godo discussion, thanks. > > > > In addition to hardware limits, also try to avoid requiring internet > > access. > > > > -- > > James Cameron > > [2]http://quozl.netrek.org/ > > ___ > > Sugar-devel mailing list > > [3]Sugar-devel@lists.sugarlabs.org > > [4]http://lists.sugarlabs.org/listinfo/sugar-devel > > > > References: > > > > [1] mailto:qu...@laptop.org > > [2] http://quozl.netrek.org/ > > [3] mailto:Sugar-devel@lists.sugarlabs.org > > [4] http://lists.sugarlabs.org/listinfo/sugar-devel > > -- > James Cameron > http://quozl.netrek.org/ > ___ > Sugar-devel mailing list > Sugar-devel@lists.sugarlabs.org > http://lists.sugarlabs.org/listinfo/sugar-devel > ___ Sugar-devel mailing list Sugar-devel@lists.sugarlabs.org http://lists.sugarlabs.org/listinfo/sugar-devel
Re: [Sugar-devel] To Cameron: ML idea discussion
As for hardware, I am not sure how strong should it be to run a training model for Neural Network or any other classifier. As for internet access, the training can be done offline no problem, it can be slow depending on the size of the data set they add, it could be a feature to try predict or estimate the amount of time it will take them given a minimum hardware specification, the problem however will be the lack of pictures to use, so a minimum number of photos and categories can be provided offline at any given time, with other pictures and categories available downloadable whenever internet is available. On Thu, 28 Mar 2019 at 02:41, James Cameron wrote: > Godo discussion, thanks. > > In addition to hardware limits, also try to avoid requiring internet > access. > > -- > James Cameron > http://quozl.netrek.org/ > ___ > Sugar-devel mailing list > Sugar-devel@lists.sugarlabs.org > http://lists.sugarlabs.org/listinfo/sugar-devel > ___ Sugar-devel mailing list Sugar-devel@lists.sugarlabs.org http://lists.sugarlabs.org/listinfo/sugar-devel
[Sugar-devel] Fwd: To Cameron: ML idea discussion
-- Forwarded message - From: Ahmed ElSabbagh Date: Wed, 27 Mar 2019 at 16:06 Subject: Re: [Sugar-devel] To Cameron: ML idea discussion To: IQRA MUHAMMAD Hi IQRA, I do understand how training and testing works, but I was discussing the way (which I had in my mind) to turn it into an interactive game. And one way I imagined was that the children will already have pictures and they teach the model which is which. In fact I just saw the links and the PDF you sent Is almost exactly the way I imagined it. But apparently it doesn't need as much images as I thought it would need. Anyway, how about that, instead of making it a fixed game of cups and cars, the game instead shows a list of words, between 2 to 8 max, a small search engine built in the activity allows the children to look for those pictures, drag and drop them to the right box. The game would then introduce other pictures for comparison, and a score is made to show how good did the children do in teaching the model. Could this count for a good idea? Or is it a bit repetitive? Regards, AHS. On Wed, 27 Mar 2019 at 12:17, IQRA MUHAMMAD wrote: > Hi! > > The ML activity was an idea given by me initially. I have to say that you > need to first understand the concept of what "testing" and "training" a > model is. We are not arranging the pictures here. We are training the > model with the pictures. > Have a look at this worksheet that I have attached along this email > and besides this ideas there are more ideas of google AI experiments that > can you get inspired from: > There is a website soley based on teaching ML to kids and that has lesson > plans and worksheets: > > 1. https://machinelearningforkids.co.uk/#!/worksheets > > 2. https://experiments.withgoogle.com/teachable-machine > > 3. https://experiments.withgoogle.com/handwriting-with-a-neural-net > > 4. https://experiments.withgoogle.com/semantris > > worksheet-carorcup.pdf > <https://drive.google.com/a/ce.ceme.edu.pk/file/d/1ZY5DdWgeSrzaswRhhXemkVWpllIV4suA/view?usp=drive_web> > Regards, > Iqra > > > On Wed, Mar 27, 2019 at 3:00 PM Ahmed ElSabbagh < > ahmed.h.elsabb...@gmail.com> wrote: > >> Hello James, >> Thank you for the warm welcome, I am currently reading the documentation >> you have sent. >> Image recognition seems interesting, but shall we fill the blanks? >> Should the activity for example show different pictures of cars, cups and >> other things, the children should then choose which picture goes to which >> category, then they test it on other images and if it should succeed in >> recognizing it? And the initially 0 knowledge model learns incrementally? >> This could be done by a score for how many new pictures the Model was >> able to recognize and each level the children pass they go to a new >> threshold or level repeat and so on. >> Problems I see here is: >> I am not familiar enough with image recognition ML field in particular, >> but shouldn't the model take many pictures in order to be capable enough? >> If so it would be tiresome for children to arrange all the pictures. >> >> On Wed, 27 Mar 2019 at 07:49, >> wrote: >> >>> Send Sugar-devel mailing list submissions to >>> sugar-devel@lists.sugarlabs.org >>> >>> To subscribe or unsubscribe via the World Wide Web, visit >>> http://lists.sugarlabs.org/listinfo/sugar-devel >>> or, via email, send a message with subject or body 'help' to >>> sugar-devel-requ...@lists.sugarlabs.org >>> >>> You can reach the person managing the list at >>> sugar-devel-ow...@lists.sugarlabs.org >>> >>> When replying, please edit your Subject line so it is more specific >>> than "Re: Contents of Sugar-devel digest..." >>> >>> >>> Today's Topics: >>> >>>1. Re: Expressing Interest GSoC (James Cameron) >>>2. Re: A problem of "No module named..." when trying to run an >>> activity which uses GTK (James Cameron) >>>3. Re: Regarding port to python 3 project (James Cameron) >>>4. Re: GSOC19 proposal (James Cameron) >>> >>> >>> -- >>> >>> Message: 1 >>> Date: Wed, 27 Mar 2019 15:56:56 +1100 >>> From: James Cameron >>> To: sugar-devel@lists.sugarlabs.org >>> Subject: Re: [Sugar-devel] Expressing Interest GSoC >>> Message-ID: <20190327045656.gn13...@laptop.org> >>> Content-Type: text/plain; charset=us-ascii >>> >>> Welcome Ahmed, >>> >>
[Sugar-devel] To Cameron: ML idea discussion
Hello James, Thank you for the warm welcome, I am currently reading the documentation you have sent. Image recognition seems interesting, but shall we fill the blanks? Should the activity for example show different pictures of cars, cups and other things, the children should then choose which picture goes to which category, then they test it on other images and if it should succeed in recognizing it? And the initially 0 knowledge model learns incrementally? This could be done by a score for how many new pictures the Model was able to recognize and each level the children pass they go to a new threshold or level repeat and so on. Problems I see here is: I am not familiar enough with image recognition ML field in particular, but shouldn't the model take many pictures in order to be capable enough? If so it would be tiresome for children to arrange all the pictures. On Wed, 27 Mar 2019 at 07:49, wrote: > Send Sugar-devel mailing list submissions to > sugar-devel@lists.sugarlabs.org > > To subscribe or unsubscribe via the World Wide Web, visit > http://lists.sugarlabs.org/listinfo/sugar-devel > or, via email, send a message with subject or body 'help' to > sugar-devel-requ...@lists.sugarlabs.org > > You can reach the person managing the list at > sugar-devel-ow...@lists.sugarlabs.org > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Sugar-devel digest..." > > > Today's Topics: > >1. Re: Expressing Interest GSoC (James Cameron) >2. Re: A problem of "No module named..." when trying to run an > activity which uses GTK (James Cameron) >3. Re: Regarding port to python 3 project (James Cameron) >4. Re: GSOC19 proposal (James Cameron) > > > -- > > Message: 1 > Date: Wed, 27 Mar 2019 15:56:56 +1100 > From: James Cameron > To: sugar-devel@lists.sugarlabs.org > Subject: Re: [Sugar-devel] Expressing Interest GSoC > Message-ID: <20190327045656.gn13...@laptop.org> > Content-Type: text/plain; charset=us-ascii > > Welcome Ahmed, > > Further information about how Sugar works for children can be found in > the documentation https://help.sugarlabs.org/ > > You can also find more about Sugarizer on the https://sugarizer.org/ > site. > > Machine Learning is difficult to fit into elementary teaching, as it > is such a narrow and rapidly changing field, with a large base of > prerequisite knowledge. > > Yes, there were ideas about Machine Learning, my previous post gave a > link to https://github.com/sugarlabs/GSoC/issues/16 > > Think of a lesson plan for how a teacher could explain Machine > Learning to a child of nine years of age, without treating it as a > "black box" concept? > > Each week I teach a class of about 12 children, and they don't need to > know about Machine Learning or Artificial Intelligence in order to use > the products that depend on it. It is often enough to say that it is > practical mathematics and logic; which is a deferral. > > On Tue, Mar 26, 2019 at 09:21:09AM +0200, Ahmed ElSabbagh wrote: > > Hello Sugar Developers, > > I am a student looking for ideas for google Google Summer of Code. > > Your SugarLabs have caught my attention but I find it difficult to > understand > > how it works in general as not many tutorials are available. > > I am fairly experienced in writing Python and I wish if possible to > create a > > new activity for Machine Learning in Sugar. > > Where could I possibly start and are there any specific ideas for ML > > applications already in your mind for implementation. > > Regards, > > AHS > > > ___ > > Sugar-devel mailing list > > Sugar-devel@lists.sugarlabs.org > > http://lists.sugarlabs.org/listinfo/sugar-devel > > > -- > James Cameron > http://quozl.netrek.org/ > > > -- > > Message: 2 > Date: Wed, 27 Mar 2019 16:02:59 +1100 > From: James Cameron > To: sugar-devel@lists.sugarlabs.org > Subject: Re: [Sugar-devel] A problem of "No module named..." when > trying to run an activity which uses GTK > Message-ID: <20190327050259.go13...@laptop.org> > Content-Type: text/plain; charset=iso-8859-1 > > Yes, Ashish Kumar made a commit which I merged as 0c53e0b, but you're > talking about a different issue, for porting to Python 3 and you've > made changes in ce7a724 ... but now I'm confused, because it doesn't > look like this commit was a complete port to Python 3, yet it is not > marked as a draft pull request. > > > https://github.com/sugarla
[Sugar-devel] Expressing Interest GSoC
Hello Sugar Developers, I am a student looking for ideas for google Google Summer of Code. Your SugarLabs have caught my attention but I find it difficult to understand how it works in general as not many tutorials are available. I am fairly experienced in writing Python and I wish if possible to create a new activity for Machine Learning in Sugar. Where could I possibly start and are there any specific ideas for ML applications already in your mind for implementation. Regards, AHS ___ Sugar-devel mailing list Sugar-devel@lists.sugarlabs.org http://lists.sugarlabs.org/listinfo/sugar-devel