- Pharo-users - lists.pharo.org

Re: [Pharo-users] Neural Networks in Pharo

by francesco agati

Hi Oleks, there is a mode for install neural network from metacello? 2017-04-25 13:00 GMT+02:00 Alexandre Bergel <alexandre.bergel(a)me.com>: > Continue to push that topic Oleks. You are on the right track! > > Alexandre > > > On Apr 24, 2017, at 1:43 AM, Oleks <olk.zaytsev(a)gmail.com> wrote: > > > > Hello, > > > > Thanks a lot for your advice! It was very helpful and educating (for > > example, I thought that we store biases in the weight matrix and prepend > 1 > > to input to make it faster, but now I see why it's actually slower that > > way). > > > > I've implemented a multi-layer neural network as a linked list of layers > > that propagate the input and error from one to another, similar to the > Chain > > of Responsibility pattern. Also, now I represent biases as separate > vectors. > > The LearningAlgorithm is a separate class with Backpropagation as its > > subclass (though at this point the network can only learn through > > backpropagation, but I'm planning to change that). I'm trying to figure > out > > how the activation and cost functions should be connected. For example, > > cross-entropy works best with logistic sigmoid activation etc. I would > like > > to give the user a freedom to use whatever he wants (plug in whatever you > > like and see what happens), but it can be very inefficient (because some > > time-consuming parts of activation and cost derivatives cancel out each > > other). > > > > Also, there is an interface for setting the learning rate for the whole > > network, which can be used to choose the learning rate prior to > learning, as > > well as to change the learning rate after each iteration. I am planning > to > > implement some optimization algorithms that would automize the process of > > choosing a learning rate (adagrad for example), but this would require a > bit > > different design (maybe I will implement the Optimizer, as you > suggested). > > > > I'm attaching two images with UML diagrams, describing my current > > implementation. Could you please tell me what you think about this > design? > > The first image is a class diagram that shows the whole architecture, and > > the second one is a sequence diagram of backpropagation. > > > > mlnn.png <http://forum.world.st/file/n4943698/mlnn.png> > > backprop.png <http://forum.world.st/file/n4943698/backprop.png> > > > > Sincerely yours, > > Oleksandr > > > > > > > > -- > > View this message in context: http://forum.world.st/Neural- > Networks-in-Pharo-tp4941271p4943698.html > > Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com. > > > > -- > _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: > Alexandre Bergel http://www.bergel.eu > ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;. > > > > >

April 25, 2017

Re: [Pharo-users] Neural Networks in Pharo

by Oleks

Hello, There isn't one yet. But I will try to create it today. I will let you know Cheers, Oleks On Apr 25, 2017 16:10, "francescoagati [via Smalltalk]" < ml+s1294792n4944025h0(a)n4.nabble.com> wrote: > Hi Oleks, > there is a mode for install neural network from metacello? > > 2017-04-25 13:00 GMT+02:00 Alexandre Bergel <[hidden email] > <http:///user/SendEmail.jtp?type=node&node=4944025&i=0>>: > >> Continue to push that topic Oleks. You are on the right track! >> >> Alexandre >> >> > On Apr 24, 2017, at 1:43 AM, Oleks <[hidden email] >> <http:///user/SendEmail.jtp?type=node&node=4944025&i=1>> wrote: >> > >> > Hello, >> > >> > Thanks a lot for your advice! It was very helpful and educating (for >> > example, I thought that we store biases in the weight matrix and >> prepend 1 >> > to input to make it faster, but now I see why it's actually slower that >> > way). >> > >> > I've implemented a multi-layer neural network as a linked list of layers >> > that propagate the input and error from one to another, similar to the >> Chain >> > of Responsibility pattern. Also, now I represent biases as separate >> vectors. >> > The LearningAlgorithm is a separate class with Backpropagation as its >> > subclass (though at this point the network can only learn through >> > backpropagation, but I'm planning to change that). I'm trying to figure >> out >> > how the activation and cost functions should be connected. For example, >> > cross-entropy works best with logistic sigmoid activation etc. I would >> like >> > to give the user a freedom to use whatever he wants (plug in whatever >> you >> > like and see what happens), but it can be very inefficient (because some >> > time-consuming parts of activation and cost derivatives cancel out each >> > other). >> > >> > Also, there is an interface for setting the learning rate for the whole >> > network, which can be used to choose the learning rate prior to >> learning, as >> > well as to change the learning rate after each iteration. I am planning >> to >> > implement some optimization algorithms that would automize the process >> of >> > choosing a learning rate (adagrad for example), but this would require >> a bit >> > different design (maybe I will implement the Optimizer, as you >> suggested). >> > >> > I'm attaching two images with UML diagrams, describing my current >> > implementation. Could you please tell me what you think about this >> design? >> > The first image is a class diagram that shows the whole architecture, >> and >> > the second one is a sequence diagram of backpropagation. >> > >> > mlnn.png <http://forum.world.st/file/n4943698/mlnn.png> >> > backprop.png <http://forum.world.st/file/n4943698/backprop.png> >> > >> > Sincerely yours, >> > Oleksandr >> > >> > >> > >> > -- >> > View this message in context: http://forum.world.st/Neural-N >> etworks-in-Pharo-tp4941271p4943698.html >> > Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com. >> > >> >> -- >> _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: >> Alexandre Bergel http://www.bergel.eu >> ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;. >> >> >> >> >> > > > ------------------------------ > If you reply to this email, your message will be added to the discussion > below: > http://forum.world.st/Neural-Networks-in-Pharo-tp4941271p4944025.html > To unsubscribe from Neural Networks in Pharo, click here > <http://forum.world.st/template/NamlServlet.jtp?macro=unsubscribe_by_code&no…> > . > NAML > <http://forum.world.st/template/NamlServlet.jtp?macro=macro_viewer&id=instan…> > -- View this message in context: http://forum.world.st/Neural-Networks-in-Pharo-tp4941271p4944027.html Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com.

April 25, 2017

ss3 and gemsource planned downtime today at 1:30 pm PDT (-0700 UTC)

by Dale Henrichs

The server hosting http://ss3.gemtalksystems.com/ and http://seaside.gemtalksystems.com/ss will be down for a couple of hours today starting at 1:30 pm PDT for maintenance. Dale

April 25, 2017

Re: [Pharo-users] Hot to retrieve values from Nested Dictionaries

by Markus BÃ¶hm

Tx, I tried Your proposal. Just as an example, it works: (ZnClient new url: 'http://api.wunderground.com/api/APIKEY/hourly/q/germany/munich.json '; contentReader: [ :entity | (NeoJSONReader on: entity readStream) mapClass: NeoJSONObject; propertyNamesAsSymbols: true; next ]; get ) hourly_forecast collect: [:e | e FCTTIME hour_padded asInteger ] P.S.: Editor in Playground shows NeoJSONObject keys in red color? Am 24.04.2017 9:00 nachm. schrieb "Udo Schneider" < udo.schneider(a)homeaddress.de>: > If you are using NeoJSON you might want to take a look at the > NeoJSONObject class. It's basically a Dictionary with "JavaScript" behavior. > I.e. you can simply use messages to access keys and non existing keys > return nil. Your example can thus be written as: > > val := dict1 key1 key2 key3. > > When using NeoJSON you can simply configure your Mapper (Reader/Writer) to > use NeoJSONObject using #mapClass: > > CU, > > Udo > > > On 24/04/2017 20:04, Markus BÃ¶hm wrote: > >> Tx for advice. >> >> P.S.: I got nested dictionaries after JSON API call at wunderground >> weather API. To see all the response in Pharo inspector "live" is really >> great :-) >> >> BR Mike >> >> >> On Mon, Apr 24, 2017 at 7:51 PM, Cyril Ferlicot D. < >> cyril.ferlicot(a)gmail.com <mailto:cyril.ferlicot@gmail.com>> wrote: >> >> Le 24/04/2017 Ã 19:42, Markus BÃ¶hm a Ã©crit : >> > May I ask: >> > What's the idiomatic way to retrieve values from nested >> dictionaries? >> > >> > (((dict1 at: 'key1') at: 'key2') at: 'key3') >> > >> > Are all the brackets really necessary? >> > >> > BR Mike >> >> Hi, >> >> To know if brackets are necessary you can use the "Format" option in >> your method. >> >> Right click -> format >> >> or cmd/alt/ctrl + shift + f >> >> In this case you can just write: >> >> ((dict1 at: 'key1') at: 'key2') at: 'key3'. >> >> >> >> -- >> Cyril Ferlicot >> https://ferlicot.fr >> >> http://www.synectique.eu >> 2 rue Jacques PrÃ©vert 01, >> 59650 Villeneuve d'ascq France >> >> >> > > >

April 25, 2017

Re: [Pharo-users] Neural Networks in Pharo

by Alexandre Bergel

Continue to push that topic Oleks. You are on the right track! Alexandre > On Apr 24, 2017, at 1:43 AM, Oleks <olk.zaytsev(a)gmail.com> wrote: > > Hello, > > Thanks a lot for your advice! It was very helpful and educating (for > example, I thought that we store biases in the weight matrix and prepend 1 > to input to make it faster, but now I see why it's actually slower that > way). > > I've implemented a multi-layer neural network as a linked list of layers > that propagate the input and error from one to another, similar to the Chain > of Responsibility pattern. Also, now I represent biases as separate vectors. > The LearningAlgorithm is a separate class with Backpropagation as its > subclass (though at this point the network can only learn through > backpropagation, but I'm planning to change that). I'm trying to figure out > how the activation and cost functions should be connected. For example, > cross-entropy works best with logistic sigmoid activation etc. I would like > to give the user a freedom to use whatever he wants (plug in whatever you > like and see what happens), but it can be very inefficient (because some > time-consuming parts of activation and cost derivatives cancel out each > other). > > Also, there is an interface for setting the learning rate for the whole > network, which can be used to choose the learning rate prior to learning, as > well as to change the learning rate after each iteration. I am planning to > implement some optimization algorithms that would automize the process of > choosing a learning rate (adagrad for example), but this would require a bit > different design (maybe I will implement the Optimizer, as you suggested). > > I'm attaching two images with UML diagrams, describing my current > implementation. Could you please tell me what you think about this design? > The first image is a class diagram that shows the whole architecture, and > the second one is a sequence diagram of backpropagation. > > mlnn.png <http://forum.world.st/file/n4943698/mlnn.png> > backprop.png <http://forum.world.st/file/n4943698/backprop.png> > > Sincerely yours, > Oleksandr > > > > -- > View this message in context: http://forum.world.st/Neural-Networks-in-Pharo-tp4941271p4943698.html > Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com. > -- _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: Alexandre Bergel http://www.bergel.eu ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.

April 25, 2017

Re: [Pharo-users] type checking in Smalltalk

by Igor Stasenko

On 31 March 2017 at 21:34, Sven Van Caekenberghe <sven(a)stfx.eu> wrote: > > > On 31 Mar 2017, at 19:38, Dimitris Chloupis <kilon.alios(a)gmail.com> > wrote: > > > > > > On Fri, 31 Mar 2017 at 19:13, Sven Van Caekenberghe <sven(a)stfx.eu> > wrote: > > if you copy/paste something you should give a reference > > > > > > I did not copy paste anything, 100 % mine. What part you think it's copy > ? > > Ah, sorry then, the formatting seemed to suggest it came from somewhere > else. > > BTW, I don't think the hybrid part is a real thing. > > Although I understand that static typed languages like C, C++, Java and so > many others have their place and use, people in those languages spend an > awful amount of time and code dealing with the types they love so much. The > modern static languages with all their magic are even worse. And even in a > nice static typed program that compiles with no warnings, there are still > dynamic errors lurking everywhere. The world cannot be defined in a static > way. > > The dynamic type errors that you get in Pharo during development are 95% > plain logic errors, once you fix those and write some unit tests, a Pharo > program is just as stable and reliable as anything else that is well > written, tested and debugged. > > And I love Igor's "Don't ask, tell" idea - right on target. > > Hey, hey.. These words, originally, definitely are not mine.. It is something i learned, being among wise people of smalltalk family. > Anyway, that is my personal opinion, I don't want to convince you. > > Sven > -- Best regards, Igor Stasenko.

April 25, 2017

Re: [Pharo-users] Hot to retrieve values from Nested Dictionaries

by Esteban A. Maringolo

2017-04-24 16:10 GMT-03:00 Ramon Leon <ramon.leon(a)allresnet.com>: > On 04/24/2017 11:16 AM, Esteban A. Maringolo wrote:> >> For these use cases it >> would be nice to have some sort of syntax sugar >> as with the cascade operator, but with a chaining operator instead. > This doesn't even require syntactic sugar, just objects, and this is a more > general problem even with a back to back chain of select/detect/reject style > things. I solved this for myself years ago by introducing what I call a pipe > (nod to unix) allowing me to chain calls in a pipeline without all the > parens with each call acting on the result value of the last call. Trivially > implemented using #doesNotUnderstand: and the cascade operator. > > ^ dict1 asPipe at: 'key1'; at: 'key2'; at: 'key3'. > > Anytime I have a bunch of back to back calls that would require a lot of > parens, I just create a pipe... > > maxRoomsAvailable > ^self findBlocks asPipe > select: [ :e | e blockDate between: actualCheckInDate and: > actualCheckInDate + (nights - 1) days ]; > detectMin: [ :e | e quantityAvailable ]; > quantityAvailable Very neat implementation. I think this simple class could come as part of the base image :) Esteban A. Maringolo

April 24, 2017

Re: [Pharo-users] Hot to retrieve values from Nested Dictionaries

by Ramon Leon

On 04/24/2017 11:16 AM, Esteban A. Maringolo wrote:> For these use cases it would be nice to have some sort of syntax sugar > as with the cascade operator, but with a chaining operator instead. This doesn't even require syntactic sugar, just objects, and this is a more general problem even with a back to back chain of select/detect/reject style things. I solved this for myself years ago by introducing what I call a pipe (nod to unix) allowing me to chain calls in a pipeline without all the parens with each call acting on the result value of the last call. Trivially implemented using #doesNotUnderstand: and the cascade operator. ^ dict1 asPipe at: 'key1'; at: 'key2'; at: 'key3'. Anytime I have a bunch of back to back calls that would require a lot of parens, I just create a pipe... maxRoomsAvailable ^self findBlocks asPipe select: [ :e | e blockDate between: actualCheckInDate and: actualCheckInDate + (nights - 1) days ]; detectMin: [ :e | e quantityAvailable ]; quantityAvailable The simplest implementation is... Object>>asPipe ^ SPipe with: self Object subclass: #SPipe instanceVariableNames: 'value' classVariableNames: '' poolDictionaries: '' category: 'PharoExtensions' privateValue: anObject value := anObject yourself ^ value yourself doesNotUnderstand: aMessage ^ value := value perform: aMessage selector withArguments: aMessage arguments SPipe class>>with: anObject ^ (self new) privateValue: anObject; yourself -- Ramon Leon

April 24, 2017

Re: [Pharo-users] Hot to retrieve values from Nested Dictionaries

by Udo Schneider

If you are using NeoJSON you might want to take a look at the NeoJSONObject class. It's basically a Dictionary with "JavaScript" behavior. I.e. you can simply use messages to access keys and non existing keys return nil. Your example can thus be written as: val := dict1 key1 key2 key3. When using NeoJSON you can simply configure your Mapper (Reader/Writer) to use NeoJSONObject using #mapClass: CU, Udo On 24/04/2017 20:04, Markus BÃ¶hm wrote: > Tx for advice. > > P.S.: I got nested dictionaries after JSON API call at wunderground > weather API. To see all the response in Pharo inspector "live" is really > great :-) > > BR Mike > > > On Mon, Apr 24, 2017 at 7:51 PM, Cyril Ferlicot D. > <cyril.ferlicot(a)gmail.com > <mailto:cyril.ferlicot@gmail.com>> wrote: > > Le 24/04/2017 Ã 19:42, Markus BÃ¶hm a Ã©crit : > > May I ask: > > What's the idiomatic way to retrieve values from nested dictionaries? > > > > (((dict1 at: 'key1') at: 'key2') at: 'key3') > > > > Are all the brackets really necessary? > > > > BR Mike > > Hi, > > To know if brackets are necessary you can use the "Format" option in > your method. > > Right click -> format > > or cmd/alt/ctrl + shift + f > > In this case you can just write: > > ((dict1 at: 'key1') at: 'key2') at: 'key3'. > > > > -- > Cyril Ferlicot > https://ferlicot.fr > > http://www.synectique.eu > 2 rue Jacques PrÃ©vert 01, > 59650 Villeneuve d'ascq France > >

April 24, 2017

Re: [Pharo-users] [Ann] Pharo Sprint App to improve coordination during Pharo Sprints

by K K Subbu

On Monday 24 April 2017 08:00 PM, Juraj Kubelka wrote: > > Please, can you send me a screenshot? Or submit the screenshot > here https://github.com/JurajKubelka/PharoSprint/issues/new Done. I think the error appears because when a login fails, the message causes a horiz scroll bar to be added to the panel pushing the fields beyond the top edge. I stumbled on this error because I tried using my fogbugz id whereas the login field needs the registered email id for login. Regards .. Subbu

April 24, 2017