nlp4j
nlp4j copied to clipboard
emorynlp
Training on nlp4j-ner model
Hi, I need to add some more dataset to pre-existing model(en-ner.xz), As it is not possible in emory nlp4j now i have trained my own model (en-sam.xz) using the files below!! sam.zip
i have used the command to train model java edu.emory.mathcs.nlp.bin.NLPTrain -mode ner -c home/config-train-sample.xml -t /home/sample-trn.tsv -d /home/sample-dev.tsv -m /home/en-sam.xz New Model was created. i need to know whether i have used correct files while training?
please help me how can i add this new model(en-sam.xz) along with en-ner.xz using config-decode-en.xml? i need to load this new model in the code(nlp4j/cli/src/main/java/edu/emory/mathcs/nlp/bin/NLPDemo.java) and test it. @jdchoi77 and team Thanks in advance.
Sorry for the late reply. The sample files would not train any good model since they are tiny. You should get the OntoNotes data from LDC and use the entire dataset to train a meaningful model:
https://catalog.ldc.upenn.edu/LDC2013T19
Please let me know if you have trouble extracting NER tags from the original OntoNotes data once you get it. Thanks.
Will OntoNotes provide me a dataset in nlp4j training format or i should create a training file in format using dataset provided?
can u please review my configuration files ? sam.zip
I am trying to train a model in NER mode.
@jdchoi77 and team Thanks in advance.
OntoNotes does not come with the format that you need. I actually made the conversion script available so please take a look at this page:
https://github.com/emorynlp/ddr/blob/master/md/conversion.md#merge
Please let me know if you have more questions. Thanks.
Is my configuration files are correct or should i make any changes????
I am getting this error while loading my new model
java.io.StreamCorruptedException: unexpected reset; recursion depth: 2
at java.io.ObjectInputStream.handleReset(ObjectInputStream.java:2049)
at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1323)
at java.io.ObjectInputStream.defaultReadFields(ObjectInputStream.java:2018)
at java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:1942)
at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:1808)
at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1353)
at java.io.ObjectInputStream.defaultReadFields(ObjectInputStream.java:2018)
at java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:1942)
at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:1808)
at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1353)
at java.io.ObjectInputStream.readObject(ObjectInputStream.java:373)
at edu.emory.mathcs.nlp.common.util.NLPUtils.getComponent(NLPUtils.java:98)
at edu.emory.mathcs.nlp.decode.AbstractNLPDecoder.init(AbstractNLPDecoder.java:120)
at edu.emory.mathcs.nlp.decode.AbstractNLPDecoder.
@jdchoi77 and team Thanks in advance.
Are you trying to decode or train? These errors are coming from the decoder. If you are trying to decode, could you send me your configuration file, input file, and command you ran?
The sample file is there only for demo and is too small to be used for training. You should feed in your own data to train NER; you can obtain a large corpus from LDC for free:
https://catalog.ldc.upenn.edu/LDC2013T19
Hello , I trained NER with large LDC corpus , but still it can't train the data but it stores the model , here i enclosed my ner_config.xml file, training and developement data , NLP4JDATA.zip
the comment line status is,
sh /home/appassembler/bin/nlptrain -mode ner -c /home/config_ner_train.xml -t /home/Output.stv -d /home/Output1.stv -m /home/NLP4JMODEL/en-sam.xz
Loading ambiguity classes Loading word clusters Loading word embeddings Loading named entity gazetteers Name not implemented for OnlineComponent. Input name - en-sam.xz will be ignored. AdaGrad Mini-batch
- Max epoch: 0
- Mini-batch: 5
- Feature cutoff: 2
- Learning rate: 0.02
- LOLS: fixed = 0, decaying rate = 0.95
- RDA: 1.0E-5 Training: 0 0: Best: 0.00, epoch = -1 Saving the model
Hi I found the problem with my config.xml file , Now it is working . Thank you .