Tutorial

[fahadshery]

Looks like I hit a jackpot for creating my own models for NER, chunking etc. A small tutorial would be ideal :)

[jwijffels]

Please look at the package vignette.

[fahadshery]

I did look at it. I am still not sure how to open the ShinyApp that comes with the package. (Sorry for being thick) and How to create a training dataset. I ran your examples successfully though

[jwijffels]

Open the file in RStudio and press on the run document button in RStudio.

[fahadshery]

I have located the annotations.Rmd installed shiny and flexdashboard packages. But when I click Run -> Run All in RStudio. It gives an error: Error in output$ui_txt <- renderText({ : object 'output' not found

[jwijffels]

I can't reproduce this error. The app runs fine with the following shiny/flexdashboard version

 installed.packages()[c("shiny", "flexdashboard"), "Version"]         
shiny flexdashboard        
"1.0.5"     "0.5.1.1"

[fahadshery]

My versions are:

> installed.packages()[c("shiny", "flexdashboard"), "Version"] shiny flexdashboard "1.1.0" "0.5.1.1" is there a way to install the similar version as yours?

[fahadshery]

installed same version as yours:devtools::install_version("shiny", version = "1.0.5", repos = "http://cran.us.r-project.org") still getting the erro

[jwijffels]

sure, get the package from cran and install it

[jwijffels]

If you can't manage to install the shiny app, you need to have/create training data in the format shown in the help of ?merge.chunkrange

[fahadshery]

It's definitely not working. I installed it on Macbook and Windows and both giving the same error at output$ui_text <- renderText({) object output not found. I always thought Shiny Apps are built with server.R and ui.R? or am I supposed to open up a different file?

[jwijffels]

works fine and tested on Windows/Mac/Linux. Run as rmarkdown::run(system.file(package = "crfsuite", "app", "annotation.Rmd"))

[fahadshery]

genius! This is why it wasn't working! Now It is up and running! it is time to build custom models... I intend to use your StarSpacy package as well. What's the key difference between the two?

[fahadshery]

I would also suggest to add 2 things in your wiki/intro/ReadMe file:

1. Add this line in the intro (rmarkdown::run(system.file(package = "crfsuite", "app", "annotation.Rmd"))) if people want to use the app :)
2. Add confusion matrix instead of table(scores$label) this will show how accurate the model is after training and how good it is performing on the test set. you just have to tweak table(scores$label) to table(train$label,scores$label)
3. Lastly, for people who can't run the app, just put an example training file. This will help them format their data in the format crfsuite expects. If you can't add an example training_set then just provide a similar format mentioned by openNLP i.e. Mike who is just 10 months old.
4. He likes to play in the lounge .
5. Again, thank you for the package. Its truely looking amazing!

[juliopiubello]

as @fahadshery some more information on how to get the data in the format of crsuites expects would be highly appreciated :)

[sirhcdlanodcm]

Just a heads up - there's a small typo in the code to run the shiny app in the vignette:

And run the app with rmarkdown::run(file = system.file(package = "crfsuite", "app", "annnotation.Rmd"))

Note the 3 n's in "annotation.Rmd". Barely worth mentioning - but for people like me who like to copy and paste without reading, it might be a nice fix ;)

Super excited to explore this package!

[jwijffels]

the crfsuite R package requires you to have a tokenised data.frame (1 row per token, in the right sequence of token occurrence) containing:

• the category you want to model in IOB format (see the vignette)
• in the columns, categorical information for each token (e.g. pos tag / is the token capitalised / position of the token in the document / lemma / first 4 letters / last 4 letters / whichever you think is relevant in explaining the category you want to model) as well as information of the surrounding (as in before/after) tokens or surrounding ngrams
• an identifier of a sequence (e.g. a document identifier)

Full copy-paste reproducible examples (aprroximately 7) of these data structures are provided in

• the package vignette: https://cran.r-project.org/web/packages/crfsuite/vignettes/crfsuite-nlp.html
• the documentation of the functions crf, predict.crf, ner_download_modeldata, crf_cbind_attributes, merge.chunkrange. See https://cran.r-project.org/web/packages/crfsuite/crfsuite.pdf

May I ask you to first look there before raising issues.