today i digged a little bit deeper in loading data for training dataset, as I has a database, not csv. Was really impressed how overloaded and overcomplicated these functions are. Supposed it should take 10-15 minutes to understand how to switch data sources for 10+y java programmer, took me much longer.
Hello all, thanks for your comments. You have been heard, future examples will go deeper into describing each step. You are welcome to ask questions here regarding the code steps. -- Tom
Looks like the library's been updated since this video, am I correct in my understanding that the call to iterations on line 45 is now maxNumLineSearchIterations and that learningRate is no longer part of the builder? Also, updater has been deprecated.
nice walkthrough in using deepLearning4j. Would be great addition if each step could be explained a bit, and more importantly: what is the training meant to be predicting exactly.. very unclear what was the objective.
I already read a lot, so I felt like I could do it in no time. Now I tried this tutorial, especially because I am not very experienced with java. It is not that I do not understand the basics like why do I need an activation function or sth similar, but what functions are there in dl4j and what does this special function do. (I tried for example looking at the documentation, trying to find another way to this RecordReaderDataSetIterator function, as I have more than just 1 label for each line in my CSV-file, but even after 3 days I am still stuck and find no way around it...)
Error:"method call expected java" Just found my mistake after quite some time desperate search. XD Hope it helps for those typing while watching. In 06:38 "File(...)" was used, in 14:02 you'll see the correct version. "new File(...)".
Looking so long at 'File' without 'new' was a very disturbing experience. Also code formatting in Idea is very fast and accessible, but idk if it exists in CE.
Hello again, I just wanted to ask if the DenseLayer is more like an input layer oder a hidden layer? And if it's not an input layer, if we do not have to configure an input layer?
This is not for beginners, I recommend these videos for people who already know ML techniques and experience with other tools like Python tools, If you are starting you should start with Statistics fundamentals and any book about ML fundamentals.
Can you please leave out unnecessary dependencies as well? There are a lot of dependencies in this project, and I don't see the necessity of it all. I want to know as bare bones as possible how to use DeepLearning4j.
batch size is the number of examples (lines in the csv) ? 50 on train 50 on test ? About the rest of the movie ... that is classical NN do DL is here applied ....
I have to say that it is weird how I see this 6 years old video, is like best Java AI related video in youtube. props for creator, Deeplearning4j Apis have changed little so this is outdated. Java AI promotion is so bad state, first I get terrible Tensor flow videos(it Java api 20% features implemented is worst possible choise for Java dev), Python it is good for it but not Java. So much of AI community has moved using Python that is even hard to find good videos for Java about subject. some small changes to made it work, if this helps someone .. RecordReader rr = new CSVRecordReader('\t'); .. RecordReader rrTest = new CSVRecordReader('\t'); Nesterovs updater = new Nesterovs(); updater.setMomentum(0.9); updater.setLearningRate(learningRate); conf.. .seed(seed).maxNumLineSearchIterations(1).optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT).updater(updater) .. .weightInit(WeightInit.XAVIER).activation(Activation.RELU) .. conf.setBackpropType(BackpropType.Standard); also i ovverride ScoreIterationListener class to print my screen .. INDArray features = t.getFeatures(); worked fine
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
![The moment we stopped understanding AI [AlexNet]](/img/tr.png)
today i digged a little bit deeper in loading data for training dataset, as I has a database, not csv. Was really impressed how overloaded and overcomplicated these functions are. Supposed it should take 10-15 minutes to understand how to switch data sources for 10+y java programmer, took me much longer.
Most soothing nn tutorial ever
Hello all, thanks for your comments. You have been heard, future examples will go deeper into describing each step.
You are welcome to ask questions here regarding the code steps.
--
Tom
Hi, i would you know if the value for one label are three, how change the initial part of the code? Thanks!
Looks like the library's been updated since this video, am I correct in my understanding that the call to iterations on line 45 is now maxNumLineSearchIterations and that learningRate is no longer part of the builder? Also, updater has been deprecated.
Could I have a link plz?
nice walkthrough in using deepLearning4j.
Would be great addition if each step could be explained a bit, and more importantly: what is the training meant to be predicting exactly.. very unclear what was the objective.
Agree. I was expecting explanation for each step instead of just simply writing the code line by line.
same here
What did you guys expect? You need to read the books to understand AI ;)
www.deeplearningbook.org/
neuralnetworksanddeeplearning.com/
I already read a lot, so I felt like I could do it in no time. Now I tried this tutorial, especially because I am not very experienced with java. It is not that I do not understand the basics like why do I need an activation function or sth similar, but what functions are there in dl4j and what does this special function do. (I tried for example looking at the documentation, trying to find another way to this RecordReaderDataSetIterator function, as I have more than just 1 label for each line in my CSV-file, but even after 3 days I am still stuck and find no way around it...)
What are hidden nodes, seed, updater, momentum, dense layer builder, iteration, weight init?
Error:"method call expected java"
Just found my mistake after quite some time desperate search. XD Hope it helps for those typing while watching.
In 06:38 "File(...)" was used, in 14:02 you'll see the correct version. "new File(...)".
Oh yeah and I really appreciate you making the video! Big thanks!
As a software developer it is incredibly painful to watch Tom type!
he probably does it to explain things better. fast forward is the way to go. he talks too slow too
@Larry Kawrence I think he's talkning about the horrid indentation.
Sadly, I can't figure out how to set up the project. It simply doesn't work for me. It seems to start from the middle of no where.
Looking so long at 'File' without 'new' was a very disturbing experience. Also code formatting in Idea is very fast and accessible, but idk if it exists in CE.
Yes, it is available in CE
Great video.
Hello again, I just wanted to ask if the DenseLayer is more like an input layer oder a hidden layer? And if it's not an input layer, if we do not have to configure an input layer?
Sir ,what is the prerequisits for learning . and What is the use of it over machine learning.
Some pattern must exist in the input for the network to learn the required output. Less feature engineering required than machine learning
Just coding without any semantic information about what he is actually doing. Very bad!!! Especially for beginners in ML.
Like... actually wtf did he do omegalul
This is not for beginners, I recommend these videos for people who already know ML techniques and experience with other tools like Python tools, If you are starting you should start with Statistics fundamentals and any book about ML fundamentals.
Can you please leave out unnecessary dependencies as well? There are a lot of dependencies in this project, and I don't see the necessity of it all. I want to know as bare bones as possible how to use DeepLearning4j.
Hi. could you post another series with the latest DL4j 1.0.0 Beta 7? I am having major problems with my code. thanks.
the website doesn't work
I clicked on it and it said 404
Nice video!
Okay, if this is outdated is there a (link to a) current one available that could be posted here?
batch size is the number of examples (lines in the csv) ? 50 on train 50 on test ?
About the rest of the movie ... that is classical NN do DL is here applied ....
Wtf was that.....I didn't learn anything!!!
github repo doesn't appear to exist?
why i'm getting an error in the FileSplit method telling me : The method File(String) is undefined for the type ... ?
He forgot to put the new in front of File, so the code should be:
new File("pathToFile.csv")
Hi,
I wanna know how to convert hand written text to binary text using java.
Thanks
Could I have a link for download?
Two weightInit calls at 11:59 in layer 1 builder.
We are writing this and then writing this .. . . what is THIS ?? So just give us the codes if you dont explain .
sir, this is for beginners or...?
no
where do i get the recordreader from
import org.datavec.api.records.reader.RecordReader;
do you need to always run mvn clean install?
I have to say that it is weird how I see this 6 years old video, is like best Java AI related video in youtube. props for creator, Deeplearning4j Apis have changed little so this is outdated. Java AI promotion is so bad state, first I get terrible Tensor flow videos(it Java api 20% features implemented is worst possible choise for Java dev), Python it is good for it but not Java. So much of AI community has moved using Python that is even hard to find good videos for Java about subject.
some small changes to made it work, if this helps someone ..
RecordReader rr = new CSVRecordReader('\t');
..
RecordReader rrTest = new CSVRecordReader('\t');
Nesterovs updater = new Nesterovs();
updater.setMomentum(0.9);
updater.setLearningRate(learningRate);
conf..
.seed(seed).maxNumLineSearchIterations(1).optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT).updater(updater)
..
.weightInit(WeightInit.XAVIER).activation(Activation.RELU)
..
conf.setBackpropType(BackpropType.Standard);
also i ovverride ScoreIterationListener class to print my screen
..
INDArray features = t.getFeatures();
worked fine
Idk why but this sound scary