zhongkaifu / rnnsharp Goto Github PK

hi, i notice that the logsumexp function in your project is not match with what's introduced in wikipedia : https://en.wikipedia.org/wiki/LogSumExp . Am I mis-understood ?

Hi,
RNNSharp is great for .net developers. but sometimes training time is kinda long.
I found one interesting project managedCuda which is complete .net wrapper for the CUDA Driver API. so maybe it's a good idea to merge it into rnnsharp.
best regards!
Adesun

Hello!
You wrote:
"You need to train an auto encoder-decoder model by RNNSharp at first, and then use this pretrained model for your task"

But I do not see how to do it.
Finaly what do you mean by "pretrained model is trained by RNNSharp itself"?

Hello Zhong,
I will be very thankful if kindly share its Pseudo code

Hello !

Thanks again for this awesome project !

From my understanding, the performance for text sequential tagging should be equal or more than CRFSharp or CRF++ right ?
My problem is to semantic tag a big single, continuous text ( hundreds of pages).

In CRF++ I get a error token ratio of about 0.5%. In RNNSharp I can´t get it better than 40%. a gigantic different. I tried LSTM and BPTT with CRF on or off. No luck.

This is expected for my use case, or am I doing something wrong ?

Decoding crashes when using dropout layer in current code version. Although works in latest demo package release version.

Unhandled Exception: System.NullReferenceException: Object reference not set to an instance of an object.
at RNNSharp.DropoutLayer.Load(BinaryReader br, LayerType layerType, Boolean forTraining) in D:\RNNSharp-master\RNNSharp\Layers\DropoutLayer.cs:line 174
at RNNSharp.Networks.RNN1.Load(LayerType layerType, BinaryReader br, Boolean forTraining) in D:\RNNSharp-master\RNNSharp\Networks\RNN.cs:line 509 at RNNSharp.Networks.BiRNN1.LoadModel(String filename, Boolean bTrain) in D:\RNNSharp-master\RNNSharp\Networks\BiRNN.cs:line 598
at RNNSharp.RNNDecoder..ctor(Config config) in D:\RNNSharp-master\RNNSharp\RNNDecoder.cs:line 24
at RNNSharpConsole.Program.Test() in D:\RNNSharp-master\RNNSharpConsole\Program.cs:line 371
at RNNSharpConsole.Program.Main(String[] args) in D:\RNNSharp-master\RNNSharpConsole\Program.cs:line 305

Hello.

info,20.02.2016 21:08:27 [TRACE] In training: log probability = -6539,9872918264
9, cross-entropy = 0,0073893168866162, perplexity = 1,00513502343797
info,20.02.2016 21:08:27
info,20.02.2016 21:08:27 Saving better model into file model.bin...
info,20.02.2016 21:08:27 Saving input2hidden weights...
info,20.02.2016 21:08:28 Saving matrix with VQ 256...
info,20.02.2016 21:10:28 Sorting data set (size: 1707219900)...

Process is terminated due to StackOverflowException.

Process is terminated due to StackOverflowException.
Process is terminated due to StackOverflowException.

Seems an error occurs in AdvUtils.dll

Hello, I was wondering what are the training times for the demonstrations.
I just tried the english seq labeler, and it took 1 hour to process 10% of the corpus! (is this normal?)
It's known Deep Learning is CPU hungry, I have only 2 cores and 8GB RAM (sorry)
¿do I need to change the PC) or acquire a CUDA core to help computing?
¿Is there a way to stop learning manually, or programmatically after reaching certain error rate?

I am wondering if you ever tried sequence labelling on highly inflectional languages (like Spanish) which has lots of inflectional power (complexity) and the words as a whole string are useless, the vocabulary explodes into >300M words! and the "examples"found on text begins to be too sparse, even with negative sampling you never get certain combinations, because most verbs have over 200 versions of itself (inflections), including time-tense, person, gender, plurality, mode, etc. So there is need to train on higher level features, but not losing the "semantic" sense. ¿do you think this could be possible, like decomposing the words (by means of controlled independent lemmatization) into parts/chunks (prefix, root, suffix, as well as modal information and semantic features of the parts,) My intuition is that this might lower the training and may be better the generalization power with less extensive corpus. Like capturing higher level syntax rules, and by the way generating semantic content constraints (may be even some common sense)...

It's just a question, on theory!

Hello!
You do a great job!
Is it possible to make incremental training?

Hello Sir,
hope you will be fine
Would you please like to let me know the impact of run-time features on results, simply why we use run-time features?

Hello,
I found when training the NER sample you have provided in the Demo package, that the model.bin file size is very large, over 1GB. What exactly is inside this file, and is there any way to reduce the size of it?
Thanks

Hi, Zhong!
Thank you for such an amazing tooling!

What do you think about publishing core libraries to Nuget. That would make life easier when it comes to updates and could be easily achieved by using free appveyor CI. I am talking about AdvUnits, Txt2Vec, CRFSharp, RNNSharp.

If you're interested, I can help you with it.

Hi,

I highly appreciate your work and thank you very much for providing this model in C#. To add in a bit of context, I will be using your model during my research, and I would like to know as much as possible about how to exploit RNNSharp for NER. I would greatly appreciate if you could answer some of my doubts and questions that I haven't seen enough explained in the README or in other issues written by other users:

1. What is validated corpus in the process of encoding a model? How does it differ from the training corpus file?

2. Is there any quicker way of preparing training files in the format required for the encoding process? Say I would like to use a CONLL training corpus but, as the format is different, manually formatting requires a huge amount of time. How to go about?

3. How to do gazzeteer-matching? Say I have a list of named entities of locations from GeoNames and I want the algorithm to integrate them and do named entity matching in the decoding process. Is there any guide how to do so?

4. I would like to know how to feed features into the algorithm, such as handcrafted linguistic features (capitalization, linguistic pattern combinations and the like) to build a hybrid linguistic and deep-learning-based NER system.

5. How to perform evaluation metrics on our test dataset such as recall, precision and f1-score?

Thank you very much in advance, and have a nice day.

I am struggling porting the library to .Net core.
Is there something to keep in mind while doing so (adding some dependencies, etc )?
Thanks for the help!

I would like to use the LSTM for sequence prediction, I enter a series of numbers, and the network, outputs what number is more likely to come next, with a percent probability. Is this possible with this library? would you have a tutorial/ example on how could something similar be achieved?

Hello!
After the last update, I get this error:

Необработанное исключение: System.AggregateException: Произошла одна или несколько ошибок. ---> System.NullReferenceException: Ссылка на объект не указывает на экземпляр объекта.
в RNNSharp.SimpleLayer.BackwardPass() в C:\Users\vanin\Documents\GitHub\RNNSharp\RNNSharp\Layers\SimpleLayer.cs:строка 288
в RNNSharp.Networks.ForwardRNN1.ProcessSequenceCRF(Sequence pSequence, RunningMode runningMode) в C:\Users\vanin\Documents\GitHub\RNNSharp\RNNSharp\Networks\FowardRNN.cs:строка 483 в RNNSharp.RNNEncoder1.Process(RNN1 rnn, DataSet1 trainingSet, RunningMode runningMode) в C:\Users\vanin\Documents\GitHub\RNNSharp\RNNSharp\RNNEncoder.cs:строка 81
в System.Threading.Tasks.Parallel.<>c__DisplayClass17_01.<ForWorker>b__1() в System.Threading.Tasks.Task.InnerInvokeWithArg(Task childTask) в System.Threading.Tasks.Task.<>c__DisplayClass176_0.<ExecuteSelfReplicating>b__0(Object ) --- Конец трассировки внутреннего стека исключений --- в System.Threading.Tasks.Task.ThrowIfExceptional(Boolean includeTaskCanceledExceptions) в System.Threading.Tasks.Task.Wait(Int32 millisecondsTimeout, CancellationToken cancellationToken) в System.Threading.Tasks.Parallel.ForWorker[TLocal](Int32 fromInclusive, Int32 toExclusive, ParallelOptions parallelOptions, Action1 body, Action2 bodyWithState, Func4 bodyWithLocal, Func1 localInit, Action1 localFinally)
в System.Threading.Tasks.Parallel.For(Int32 fromInclusive, Int32 toExclusive, Action1 body) в RNNSharp.RNNEncoder1.Train() в C:\Users\vanin\Documents\GitHub\RNNSharp\RNNSharp\RNNEncoder.cs:строка 227
в RNNSharpConsole.Program.Train() в C:\Users\vanin\Documents\GitHub\RNNSharp\RNNSharpConsole\Program.cs:строка 550
в RNNSharpConsole.Program.Main(String[] args) в C:\Users\vanin\Documents\GitHub\RNNSharp\RNNSharpConsole\Program.cs:строка 310`

before update it is working without problems.

My config is :

`CURRENT_DIRECTORY = .\model2cfg

MODEL_TYPE = SEQLABEL

#Forward and BiDirectional are supported
NETWORK_TYPE = Forward

MODEL_FILEPATH = model.bin

HIDDEN_LAYER = LSTM:200, LSTM:100

OUTPUT_LAYER = Simple

CRF_LAYER = True

TFEATURE_FILENAME = tfeatures

#Binary and Freq are supported
TFEATURE_WEIGHT_TYPE = Freq

TFEATURE_CONTEXT = -7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5

PRETRAIN_TYPE = Embedding

WORDEMBEDDING_FILENAME = vector_big.bin

WORDEMBEDDING_CONTEXT = -1,0,1

WORDEMBEDDING_COLUMN = 0

AUTOENCODER_CONFIG = D:\RNNSharpDemoPackage\config_autoencoder.txt

#SEQ2SEQ_AUTOENCODER_CONFIG: D:\RNNSharpDemoPackage\config_seq2seq_autoencoder.txt

RTFEATURE_CONTEXT = -2,-1`

'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_64\mscorlib\v4.0_4.0.0.0__b77a5c561934e089\mscorlib.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\assembly\GAC_MSIL\Microsoft.VisualStudio.HostingProcess.Utilities\14.0.0.0__b03f5f7f11d50a3a\Microsoft.VisualStudio.HostingProcess.Utilities.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_MSIL\System.Windows.Forms\v4.0_4.0.0.0__b77a5c561934e089\System.Windows.Forms.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_MSIL\System\v4.0_4.0.0.0__b77a5c561934e089\System.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_MSIL\System.Drawing\v4.0_4.0.0.0__b03f5f7f11d50a3a\System.Drawing.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\assembly\GAC_MSIL\Microsoft.VisualStudio.HostingProcess.Utilities.Sync\14.0.0.0__b03f5f7f11d50a3a\Microsoft.VisualStudio.HostingProcess.Utilities.Sync.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\assembly\GAC_MSIL\Microsoft.VisualStudio.Debugger.Runtime\14.0.0.0__b03f5f7f11d50a3a\Microsoft.VisualStudio.Debugger.Runtime.dll'.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\Users\Aguilar\Desktop\RNNSharp-master\RNNSharpConsole\bin\Debug\RNNSharpConsole.vshost.exe'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_MSIL\System.Core\v4.0_4.0.0.0__b77a5c561934e089\System.Core.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_MSIL\System.Xml.Linq\v4.0_4.0.0.0__b77a5c561934e089\System.Xml.Linq.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_MSIL\System.Data.DataSetExtensions\v4.0_4.0.0.0__b77a5c561934e089\System.Data.DataSetExtensions.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_MSIL\Microsoft.CSharp\v4.0_4.0.0.0__b03f5f7f11d50a3a\Microsoft.CSharp.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_64\System.Data\v4.0_4.0.0.0__b77a5c561934e089\System.Data.dll'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\WINDOWS\Microsoft.Net\assembly\GAC_MSIL\System.Xml\v4.0_4.0.0.0__b77a5c561934e089\System.Xml.dll'. Symbols loaded.
The thread 0x4a7a4 has exited with code 0 (0x0).
The thread 0x4a570 has exited with code 0 (0x0).
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\Users\Aguilar\Desktop\RNNSharp-master\RNNSharpConsole\bin\Debug\RNNSharpConsole.exe'. Symbols loaded.
'RNNSharpConsole.vshost.exe' (CLR v4.0.30319: RNNSharpConsole.vshost.exe): Loaded 'C:\Users\Aguilar\Desktop\RNNSharp-master\RNNSharpConsole\bin\Debug\AdvUtils.dll'. Cannot find or open the PDB file.
The thread 0x4a464 has exited with code 0 (0x0).
The thread 0x19050 has exited with code 0 (0x0).
The program '[292464] RNNSharpConsole.vshost.exe' has exited with code 0 (0x0).

AdvUtils.dll is in that folder but it says cannot find or open. Any suggestions?

Hi, Zhong! Thank you for your lib, your work looks really cool. I've successfully run the demo, but when I change my own corpus for NER, there some errors happen as follows:

未经处理的异常: System.IndexOutOfRangeException: 索引超出了数组界限。
在 RNNSharp.RNN.ForwardBackward(Int32 numStates, Double[][] m_RawOutput)
在 RNNSharp.RNN.learnSentenceForRNNCRF(Sequence pSequence)
在 RNNSharp.RNN.TrainNet()
在 RNNSharp.RNNEncoder.Train()
在 RNNSharpConsole.Program.Main(String[] args)

Is there any limit to the lengh of sentences? (P.S. the sentences in my task is very long for each sentence containing all words in a document) Have you handled Vanishing gradient problem by some manipulation such as mini-batch like this code http://deeplearning.net/tutorial/rnnslu.html#dataset

Hello again!
Using today's version ( 4fad1b6 ), It can't converge using LTSM compared to yesterday version, and give me a crash when using the RNN.

The invoked command line was:

-mode train -trainfile wnnsharp-data-rsysyi.txt -modelfile rnnsharp.model -validfile wnnsharp-data-rsysyi.txt -ftrfile wnnsharp-config-qxnigh.txt -tagfile avaliable-tags.txt -modeltype 0 -layersize 50 -alpha 0.1 -crf 0 -maxiter 30 -savestep 200K -dir 0 -dropout 0

The error is:

Unhandled Exception: System.AggregateException: One or more errors occurred. ---> System.IndexOutOfRangeException: Index was outside the bounds of the array.
   at RNNSharp.RNN.<>c__DisplayClass107_0.<matrixXvectorADD>b__0(Int32 i) in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharp\RNN.cs:line 660
   at System.Threading.Tasks.Parallel.<>c__DisplayClass17_0`1.<ForWorker>b__1()
   at System.Threading.Tasks.Task.InnerInvokeWithArg(Task childTask)
   at System.Threading.Tasks.Task.<>c__DisplayClass176_0.<ExecuteSelfReplicating>b__0(Object )
   --- End of inner exception stack trace ---
   at System.Threading.Tasks.Task.ThrowIfExceptional(Boolean includeTaskCanceledExceptions)
   at System.Threading.Tasks.Task.Wait(Int32 millisecondsTimeout, CancellationToken cancellationToken)
   at System.Threading.Tasks.Parallel.ForWorker[TLocal](Int32 fromInclusive, Int32 toExclusive, ParallelOptions parallelOptions, Action`1 body, Action`2 bodyWithState, Func`4 bodyWithLocal, Func`1 localInit, Action`1 localFinally)
   at System.Threading.Tasks.Parallel.For(Int32 fromInclusive, Int32 toExclusive, ParallelOptions parallelOptions, Action`1 body)
   at RNNSharp.RNN.matrixXvectorADD(SimpleLayer dest, SimpleLayer srcvec, Matrix`1 srcmatrix, Int32 DestSize, Int32 SrcSize, Int32 type) in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharp\RNN.cs:line 685
   at RNNSharp.SimpleRNN.computeHiddenLayer(State state, Boolean isTrain) in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharp\SimpleRNN.cs:line 157
   at RNNSharp.RNN.PredictSentence(Sequence pSequence, RunningMode runningMode) in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharp\RNN.cs:line 267
   at RNNSharp.RNN.TrainNet(DataSet trainingSet, Int32 iter) in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharp\RNN.cs:line 536
   at RNNSharp.RNNEncoder.Train() in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharp\RNNEncoder.cs:line 133
   at RNNSharpConsole.Program.Main(String[] args) in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharpConsole\Program.cs:line 284

Hello
Hope that you will be fine.
Can you briefly explain impact of each feature type specially the embedding feature and runtime feature on the final results of RNN compared to the traditional CRF which use only one feature type e.g. the template feature.
thanks in advance

Hello !

Testing this last commit, I get a crash in UpdateLearningRate

Unhandled Exception: System.AggregateException: One or more errors occurred. ---> System.AggregateException: One or more errors occurred. ---> System.IndexOutOfRangeException: Index was outside the bounds of the array.
   at RNNSharp.RNN.UpdateLearningRate(Matrix`1 m, Int32 i, Int32 j, Double delta) in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharp\RNN.cs:line 101
at RNNSharp.LSTMRNN.<LearnOutputWeight>b__25_0(Int32 i) in C:\sbuild\mine\backend_broka\Candidatos\RNNSharp-master\RNNSharp\LSTMRNN.cs:line 609

In LSTMRNN , Line 601 do this:

Parallel.For(0, L1, parallelOption, i =>
            {
                double cellOutput = neuHidden[i].cellOutput;
                for (int k = 0; k < L2; k++)
                {
                    double delta = NormalizeGradient(cellOutput * OutputLayer.er[k]);
                    double newLearningRate = UpdateLearningRate(Hidden2OutputWeightLearningRate, i, k, delta);

                    Hidden2OutputWeight[k][i] += newLearningRate * delta;
                }
            });

But in line 517 , it is initialized like :
Hidden2OutputWeightLearningRate = new Matrix<float>(L2, L1);

when i training a pos tagger using RNN,CRF,RNN+CRF,LSTM+CRF,the result is very good, but LSTM is not better than others,I have same training, development and test corpus, what is my problem?

Hello again @zhongkaifu ,

Testing the latest version, after training, if I try to tag a file, I get the following crash:

Without WORDEMBEDDING_FILENAME , it works nice. The emmbed was created in Tex2Vec with the following command line:
Txt2VecConsole.exe -mode train -trainfile corpus.txt -modelfile vector.bin -vocabfile vocab.txt -vector-size 200 -window 5 -min-count 1 -cbow 1 -threads 1 -iter 15

info,4/8/2016 12:58:28 AM Loading template feature set...
info,4/8/2016 12:58:28 AM Template feature size: 86693
info,4/8/2016 12:58:28 AM Template feature context size: 86693
info,4/8/2016 12:58:28 AM Get model type LSTM and direction FORWARD
info,4/8/2016 12:58:28 AM Model Structure: LSTM-RNN
info,4/8/2016 12:58:28 AM Loading LSTM-RNN model: E:\rnnsharp.model
info,4/8/2016 12:58:28 AM Loading input2hidden weights...
info,4/8/2016 12:58:28 AM Loading LSTM-Weight: width:200, height:86693, vqSize:0...
info,4/8/2016 12:58:30 AM Loading feature2hidden weights...
info,4/8/2016 12:58:30 AM Loading LSTM-Weight: width:200, height:200, vqSize:0...
info,4/8/2016 12:58:30 AM Loading hidden2output weights...
info,4/8/2016 12:58:30 AM Loading matrix. width: 200, height: 22, vqSize: 0
info,4/8/2016 12:58:30 AM CRF Model: False
Unhandled Exception: System.AggregateException: One or more errors occurred. ---> System.NullReferenceException: Object reference not set to an instance of an object.
   at RNNSharp.SingleVector.get_Item(Int32 i) in C:\RNNSharp-master\RNNSharp\Vector.cs:line 107
   at RNNSharp.LSTMRNN.<>c__DisplayClass37_0.<computeHiddenLayer>b__0(Int32 j) in C:\RNNSharp-master\RNNSharp\LSTMRNN.cs:line 820
   at System.Threading.Tasks.Parallel.<>c__DisplayClass17_0`1.<ForWorker>b__1()
   at System.Threading.Tasks.Task.InnerInvokeWithArg(Task childTask)
   at System.Threading.Tasks.Task.<>c__DisplayClass176_0.<ExecuteSelfReplicating>b__0(Object )
   --- End of inner exception stack trace ---
   at System.Threading.Tasks.Task.ThrowIfExceptional(Boolean includeTaskCanceledExceptions)
   at System.Threading.Tasks.Task.Wait(Int32 millisecondsTimeout, CancellationToken cancellationToken)
   at System.Threading.Tasks.Parallel.ForWorker[TLocal](Int32 fromInclusive, Int32 toExclusive, ParallelOptions parallelOptions, Action`1 body, Action`2 bodyWithState, Func`4 bodyWithLocal, Func`1 localInit, Action`1 localFinally)
   at System.Threading.Tasks.Parallel.For(Int32 fromInclusive, Int32 toExclusive, ParallelOptions parallelOptions, Action`1 body)
   at RNNSharp.LSTMRNN.computeHiddenLayer(State state, Boolean isTrain) in C:\RNNSharp-master\RNNSharp\LSTMRNN.cs:line 869
   at RNNSharp.RNN.PredictSentence(Sequence pSequence, RunningMode runningMode) in C:\RNNSharp-master\RNNSharp\RNN.cs:line 277
   at RNNSharp.RNNDecoder.Process(Sentence sent) in C:\RNNSharp-master\RNNSharp\RNNDecoder.cs:line 77
   at RNNSharpConsole.Program.Test() in C:\RNNSharp-master\RNNSharpConsole\Program.cs:line 372
   at RNNSharpConsole.Program.Main(String[] args) in C:\RNNSharp-master\RNNSharpConsole\Program.cs:line 289


Process finished with exit code -1

Hi, does this library support training on GPU?

Hello,

I have a spatio-temporal dataset that I have compiled. It's in a TSV format, and I'd like your RNNSharp to consume the input for classification as well as recognition. My features are continuous values in the range [0, 1]. My TSV file looks like the following:

ID1 0.923 0.223 0.573 0.235 0.111
ID1 0.920 0.228 0.353 0.213 0.098
ID1 0.901 0.677 0.235 0.551 0.121
...
ID1 0.853 0.383 0.301 0.618 0.132

ID1 0.918 0.733 0.622 0.222 0.238
ID1 0.985 0.682 0.793 0.221 0.465
...
ID1 0.953 0.788 0.912 0.228 0.539

ID2 0.918 0.733 0.622 0.222 0.238
ID2 0.985 0.682 0.793 0.221 0.465
...
ID2 0.953 0.788 0.912 0.228 0.539

Each line in my TSV is a snapshot at a specific moment in time. When all snapshot are combined, it describes the spatio-temporal entity. These entities are separated by an EMPTY LINE. Therefore, the first instance ID1 is all the lines until you reach the empty line. The second instance of ID1 is the next set of contiguous lines and so on. Note, the first TSV value is just a class label and is not a feature. Also, I have 6 class labels for this spatio-temporal dataset.

1.) First, how can I transform my data into an "embedded feature" that is in the correct model format? I assume this is the Txt2Vec?

2.) Additionally, I will have to create a corpus. Will the following work for the corpus?

ID1 ClassLabel1
ID2 ClassLabel2
ID3 ClassLabel3
ID4 ClassLabel4
ID5 ClassLabel5
ID6 ClassLabel6

3.) Additional steps or a walkthrough would be greatly appreciated. I hope this information helps all others who are trying to consume RNNSharp. When I finish, I hope to compile a walkthrough for others, so they can easily consume this great technology.

Thank you.

Though RNN support the word embedding feature which is very plus point of RNN compared to the competitor CRF. is RNN have the capability to support external Gazetteer list and dictionaries as feature?

Hi, I am really impressed by your readable code. A simple question about the code which may not worth mention: I think it should stop the learning when the PPL on the development set are not increased any more or the increased error rate meets our requirement, rather than on the training set. If current ppl is larger than the previous one, we should increase the learning rate or make some other decisions.

Hello,
Testing RNNSharp I was unable to make the model converge. No matter what setting I used.
I changed two places and got the model to finally converge:

if (ppl >= lastPPL && lastAlpha != rnn.LearningRate)
 {
  //Although we reduce alpha value, we still cannot get better result.

I changed it to break only after we tried to lower alpha 8 times and failed to get a improvement.

I also changed this fragment:
rnn.LearningRate = rnn.LearningRate / 2.0f;

To a decrease in a lower rate, in my case, 1.4.

Do any of those changes make sense ?

Do you think that a smarter learning rate annealing could improve RNNSharp, or am I looking at the wrong place ??

Thanks !!

Hello Zhong,
I will be very thankful if kindly share its exact time complexity

Hello!
When I try to run RNNSharpConsole with the parameters:

MODEL_TYPE = SEQLABEL

MODEL_DIRECTION = Forward

it throws exception "NullReferenceException"

220 line of code er += srcvec[j] * srcmatrix[j][i];
file RNNHelpers.cs

I tried to solve it, but I could not understand, why DenseWeights are not initialized.

''Необработанное исключение: System.AggregateException: Произошла одна или несколь
ко ошибок. ---> System.NullReferenceException: Ссылка на объект не указывает на
экземпляр объекта.
в RNNSharp.RNNHelper.<>c__DisplayClass27_0.b__0(Int32 i)
в C:\Users\vanin\Documents\Visual Studio 2015\Projects\RNNSharpnew\RNNSharp\RNN
Helper.cs:строка 220
в System.Threading.Tasks.Parallel.<>c__DisplayClass17_01.<ForWorker>b__1() в System.Threading.Tasks.Task.InnerInvokeWithArg(Task childTask) в System.Threading.Tasks.Task.<>c__DisplayClass176_0.<ExecuteSelfReplicating> b__0(Object ) --- Конец трассировки внутреннего стека исключений --- в System.Threading.Tasks.Task.ThrowIfExceptional(Boolean includeTaskCanceledE xceptions) в System.Threading.Tasks.Task.Wait(Int32 millisecondsTimeout, CancellationTok en cancellationToken) в System.Threading.Tasks.Parallel.ForWorker[TLocal](Int32 fromInclusive, Int3 2 toExclusive, ParallelOptions parallelOptions, Action1 body, Action2 bodyWith State, Func4 bodyWithLocal, Func1 localInit, Action1 localFinally)
в System.Threading.Tasks.Parallel.For(Int32 fromInclusive, Int32 toExclusive,
Action1 body) в RNNSharp.RNNHelper.matrixXvectorADDErr(Single[] dest, Single[] srcvec, Matr ix1 srcmatrix, Int32 DestSize, Int32 SrcSize) в C:\Users\vanin\Documents\Visual
Studio 2015\Projects\RNNSharpnew\RNNSharp\RNNHelper.cs:строка 215
в RNNSharp.SimpleLayer.ComputeLayerErr(SimpleLayer nextLayer) в C:\Users\vani
n\Documents\Visual Studio 2015\Projects\RNNSharpnew\RNNSharp\SimpleLayer.cs:стро
ка 287
в RNNSharp.DropoutLayer.ComputeLayerErr(SimpleLayer nextLayer) в C:\Users\van
in\Documents\Visual Studio 2015\Projects\RNNSharpnew\RNNSharp\DropoutLayer.cs:ст
рока 124
в RNNSharp.ForwardRNN1.ProcessSequenceCRF(Sequence pSequence, RunningMode ru nningMode) в C:\Users\vanin\Documents\Visual Studio 2015\Projects\RNNSharpnew\RN NSharp\FowardRNN.cs:строка 436 в RNNSharp.RNN1.TrainNet(DataSet1 trainingSet, Int32 iter) в C:\Users\vanin \Documents\Visual Studio 2015\Projects\RNNSharpnew\RNNSharp\RNN.cs:строка 317 в RNNSharp.RNNEncoder1.Train() в C:\Users\vanin\Documents\Visual Studio 2015
\Projects\RNNSharpnew\RNNSharp\RNNEncoder.cs:строка 96
в RNNSharpConsole.Program.Train() в C:\Users\vanin\Documents\Visual Studio 20
15\Projects\RNNSharpnew\RNNSharpConsole\Program.cs:строка 542
в RNNSharpConsole.Program.Main(String[] args) в C:\Users\vanin\Documents\Visu
al Studio 2015\Projects\RNNSharpnew\RNNSharpConsole\Program.cs:строка 292

Hello zhongkaifu,

Your framework is very great but there are something I think you should change a little to get your code more easy to use. Can you change your code in the way that it can receive directly a matrix, a jagged array as input, have a train and predict method like here ? (The example)

http://accord-framework.net/docs/html/T_Accord_MachineLearning_VectorMachines_MulticlassSupportVectorMachine.htm

With this change, I think you will make a big milestone.

Regards

Hi there,

recently I wanna do some tests about sequence labelling(OCR without segment) via RNN

I googled and found this project , Thanks for your efforts

I have hundreds of handwritten word images and the corresponding word

Would you like to give me some instruction about this problem

any advice will be welcomeed. thanks in advance !

Best regards,

Hi,

Do you have a version available for VS2010 and .NET 4.0?

Hello,
i am little bit confused about the various combinations of model types and possible direction e.g forward and bidirectional. i am new to RNN architecture so please guide urgently.
is the following combinations are valid ? if not then please explain little bit. thanks in advance.
With Single Hidden Layer e.g –layersize 200
Bi-Directional LSTM-RNN without CRF
Bi-Directional LSTM-RNN with CRF
Bi-Directional BPTT-RNN without CRF
Bi-Directional BPTT-RNN with CRF
Forward LSTM-RNN without CRF
Forward LSTM-RNN with CRF
Forward BPTT-RNN without CRF
Forward BPTT-RNN with CRF
With Two Hidden Layer (deep layers) e.g –layersize 200,100
Bi-Directional LSTM-RNN without CRF
Bi-Directional LSTM-RNN with CRF
Bi-Directional BPTT-RNN without CRF
Bi-Directional BPTT-RNN with CRF
Forward LSTM-RNN without CRF
Forward LSTM-RNN with CRF
Forward BPTT-RNN without CRF
Forward BPTT-RNN with CRF

@zhongkaifu , You're awesome !! I'm very excited to lest the deep network.

However, in the latest commit, it reference the file BPTTLayer.cs. And I can't find this file in the actual git.

Hello !

First of all, thank you for publishing such high quality project. It is the absolute state-of-art ! Thank you !!!

I was evaluating this in a small dataset, however I get this ArgumentOutOfRangeException.

The error is in SimpleRNN.cs, at this part:

Logger.WriteLine("Saving feature2hidden weights...");
saveMatrixBin(mat_feature2hidden, fo);

This matrix, mat_feature2hidden, have the Height of 200, however the Width is 0.

The command:
.\Bin\RNNSharpConsole.exe -mode train -trainfile bruno-data.txt -modelfile .\bruno-model.bin -validfile bruno-valid.txt -ftrfile .\config_bruno.txt -tagfile .\bruno-tags.txt -modeltype 0 -layersize 200 -alpha 0.1 -crf 1 -maxiter 0 -savestep 200K -dir 1 -dropout 0

The config:
#The file name for template feature set
TFEATURE_FILENAME:.\bruno-features
#The context range for template feature set. In below, the context is current token, next token and next after next token
TFEATURE_CONTEXT: 0

And the error:

info,11/02/2016 14:25:38 Distortion: 0,0873744581246124, vqSize: 256
info,11/02/2016 14:25:38 Saving feature2hidden weights...
info,11/02/2016 14:25:38 Saving matrix with VQ 256...
info,11/02/2016 14:25:38 Sorting data set (size: 0)...

Unhandled Exception: System.ArgumentOutOfRangeException: Index was out of range. Must be non-negative and less than the size of the collection.
Parameter name: index
   at System.ThrowHelper.ThrowArgumentOutOfRangeException(ExceptionArgument argument, ExceptionResource resource)
   at AdvUtils.VarBigArray`1.get_Item(Int64 offset)
   at AdvUtils.VectorQuantization.BuildCodebook(Int32 vqSize)
   at RNNSharp.RNN.saveMatrixBin(Matrix`1 mat, BinaryWriter fo, Boolean BuildVQ) in C:\github\RNNSharp\RNNSharp\RNN.cs:line 138
   at RNNSharp.SimpleRNN.saveNetBin(String filename) in C:\github\RNNSharp\RNNSharp\SimpleRNN.cs:line 517
   at RNNSharp.BiRNN.saveNetBin(String filename) in C:\github\RNNSharp\RNNSharp\BiRNN.cs:line 499
   at RNNSharp.RNNEncoder.Train() in C:\github\RNNSharp\RNNSharp\RNNEncoder.cs:line 107
   at RNNSharpConsole.Program.Main(String[] args) in C:\github\RNNSharp\RNNSharpConsole\Program.cs:line 279

Maybe it is because I'm not using a word embedding ??

Thank you again for this great project !

Is there any runnable example? I've change the code a little bit and tried to run it, but the error rate is unreasonable high

Hi! Thank you for your lib, it look's really cool. But i've got problems in understanding some concepts and approaches. For example what for is feature templating? How does all those POS parameters means (i dont get the meaning of those abbreviations)
I want to test your tool with document classifications, what do i need to do in general? Or is there any tutorialt to start with?
Sorry if my questions sounds really stupid

Hi, I trained the sample, for english BIO NER labeler, adn the training was OK
But when running the TEST batch, it fails TOTALLY!
The labels got were absolutely NUTS!

so I started to din inside algorithms, and found a BIG PROBLEM
when you write down the training and model files, and use a "string" as double, in "en-US" culture, the decimal point is a "." while in spanish culture, it is a "," so when writing down and reading agfain the files, there is a inconsistency (if used as string) so the solution is 2 ways

1. use this way (modelReader.cs)

            //读入cost_factor
            strLine = sr.ReadLine();
            cost_factor_ = double.Parse(strLine.Split(':')[1].Trim(), CultureInfo.InvariantCulture);
   

2. add this to the first run inside a MAIN() loop for console apps.

   static void Main(string[] args)
    {
    	CultureInfo culture = CultureInfo.CreateSpecificCulture("en-US");
    	CultureInfo.DefaultThreadCurrentCulture = culture;
    	CultureInfo.DefaultThreadCurrentUICulture = culture;
    	Thread.CurrentThread.CurrentCulture = culture;
    	Thread.CurrentThread.CurrentUICulture = culture;
     .....
   

And now it works!

Thanks for this greet job!
I have got a problem when i start to use the new version( I used the old version on codeplex before)
it throws outofmemory exception when I used old version text2vec tool generated vector.bin file.if there are a way to convert it to new version.

WordEMWrapFeaturizer funcion
model.LoadBinaryModel(filename);
i found that there is no
bw.Write(0);//no VQ
in the old version,maybe this rises the exception,is there any solutions for this,I donot want to retrain the word_vector.bin.

Hello,

I'm wandering whether RNNSharp can be used for the following problem:

I have a sequence of 3-number vectors {[a1, a2, a3], [b1, b2, b3]... [n1, n2, n3]}. and a matching output, which is a vector of 5 numbers [A1, A2, A3, A4, A5]. And my training input/output would be a list of such pairs of sequences, where output always has the same size (5-number vector) while input size varies (number of 3-number vectors is different).

So, for example:

[1, 2, 3], [4, 5, 6] -> [1, 2, 3, 4, 5]
[1, 2, 3], [4, 5, 6], [7, 8, 9,] -> [5, 6, 7, 8, 9]
[1, 2, 3], [4, 5, 6], [7, 8, 9,], [10, 11, 12], [13, 14, 15] -> [10, 11, 12, 13, 14]
....

and I would use the samples above to train the network so that it can output a 5-number vector for a random-size input (random number of 3-number vectors)

e.g.

[1, 2, 3], [4, 5, 6], [7, 8, 9,], [10, 11, 12] -> desired 5-number output

Can RNNSharp deal with such a problem? I'm not a novice with NN, but I've never used RNN before, so I'm not sure whether it can be used with a (relative) success here. Also, if RNNSharp is capable of dealing with such a problem, any help of how can I start / setup the network would be a greatly appreciated:) (as I would be using the API in my own project)

Thanks in advance

how about the inference time for sentences up to 30 words?

Hello, First i really appreciate your work. My programming skills in C# is not good therefore apology in advance if question seems to your good-self stupid.
when i copile the source i got the error in ModelSetting .cs
System.NotSupportedException was unhandled
HResult=-2146233067
Message=Vector.Count cannot be called via reflection when intrinsics are enabled.
Source=System.Numerics.Vectors
StackTrace:
at System.Numerics.Vector`1.get_Count()
at RNNSharp.ModelSetting.DumpSetting() in D:\RNNSharp-master\RNNSharp\ModelSetting.cs:line 51
at RNNSharpConsole.Program.Train() in D:\RNNSharp-master\RNNSharpConsole\Program.cs:line 444
at RNNSharpConsole.Program.Main(String[] args) in D:\RNNSharp-master\RNNSharpConsole\Program.cs:line 277
at System.AppDomain._nExecuteAssembly(RuntimeAssembly assembly, String[] args)
at System.AppDomain.ExecuteAssembly(String assemblyFile, Evidence assemblySecurity, String[] args)
at Microsoft.VisualStudio.HostingProcess.HostProc.RunUsersAssembly()
at System.Threading.ExecutionContext.RunInternal(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
at System.Threading.ExecutionContext.Run(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
at System.Threading.ExecutionContext.Run(ExecutionContext executionContext, ContextCallback callback, Object state)
at System.Threading.ThreadHelper.ThreadStart()
InnerException:
here is my screen shoot:

so please help , i am new to C# and my work is pending since two days due to this bug. thanks

Hello ,
Can i represent DRNN in the below structure for Named entity task. any suggestion for improvement will be highly appreciated. Also is it based on Elman Architecture which is described in (Kaisheng Yao et al.,2013) ?