i got error The
nlp object should have a pretrained ner component. after train linked model
how can fix it ?
?
This code specifically, doesn't render the arc labels as shown in the video.
the code is entities=[(i, i.label_, i.start_char, i.end_char, i.sent) for i in nytimes.ents]
hi friends, I was trying to find an example from Matt Honnibal I'd seen online a while back, it involved streaming through a text corpus that also had metadata. the technique had something to do with splitting the generator so one could stream through the text and the other could yield the metadata and then joining them back up again (i.e. I want to pass one to a tokenizer and one to something that pulls out labels). this might have been in the text classification demo maybe? ringing any bells for anyone?
it was something along these lines, but with more explanation https://github.com/explosion/spaCy/blob/de69bc65098169a3f7098353d790a21b06d01f04/spacy/language.py#L756
I would be happy if anyone can help me with the response
tnx
I've got GPT2 writing some good poetry for me, and I'm trying to get another network to select the "best" of the output; wondering if I should ask about that here, or in a more general NLP community?
Dear all, I was wondering if it's possible to add additional properties to a single Span instance. For example, I extract entities from a document and I have their coordinates, which I would like to attach to the span when I map the text, tokens and entities into Spacy. Any idea on how to do that?
I checked the set/get extension but they are at class level if I understood correctly
@kakatofoli_gitlab your code works!
@lfoppiano you can set custom extensions on Token, Span, and/or Doc objects. I found this article to the most illuminating https://medium.com/@ashiqgiga07/spacy-and-the-set-extension-attributes-47a094c921c7
@galtay not sure if its what you have in mind but https://support.prodi.gy/t/how-to-incorporate-document-metadata/296/3
super curious about this https://github.com/explosion/spacy-transformers/blob/master/spacy_transformers/pipeline/ner.py anyone know how it's coming along?
I also started looking at poetry recommendation engines, and will keep researching that, bc people are coming up with lists of features to use, but those features are often quite silly/arbitrary
@delta_ark_twitter interesting application you have there. have you tried a synonym mapping (e.g. from a thesaurus or even a rhyming dictionary) for generating new combinations? just a first thought :) also, I bet adding a syllable-count attribute to your tokens would be instrumental.
@jack-rory-staunton hey Jack; doing generation is not really the problem; I can do all sorts of cool generative tricks with GPT2. The problem is discrimination. Some of what the model spits out is bad, some is so so so, and some is truly excellent. I want to be able to generate a large text (on a topic) <-- this part isn't hard, but I want another model to come in and identiy/clip the best parts and hand it to me, so I can use that to build my poem. (One day, would be nice to add a scraping component to this whole system, but making the editor/discriminator is more important now). And yeah, this is like, what I'm doing when I'm unemployed, lol.
@delta_ark_twitter i see- so its text classification then. clustering on text features ("authorship style") from e.g. textacy stats might help. binary classification (albeit requiring manual annotation) w vowpal wabbit might help too. another thing that could be cool: writing "parallel" lines by doing some word vector arithmetic a la [wv('king') - wv('queen') ~= wv('man') - wv('woman')]
@delta_ark_twitter I don't yet have much experience w text classification per se but this example is where I would start https://www.kaggle.com/poonaml/text-classification-using-spacy
Hey all! I was currently looking at training spacy with custom semantics ala https://spacy.io/usage/training#intent-parser - I was wondering - how should I go about training documents that have potentially multiple roots? eg.
find hotels with wifi and email them to susan
(perhaps not a perfect example since you could argue that emailing them, etc is connected to find... so perhaps
find stores that are open and find hotels with wifi)
Hi everyone, I have a question, that I couldn't find answer anywhere for.
So I'm training a Spacy NER model and on every iteration I see precision, recall and F1 score (as
My question is, are these values micro- or macro-averages, or something else completely?
Thanks!
So I'm training a Spacy NER model and on every iteration I see precision, recall and F1 score (as
NER P, NER R and NER F).My question is, are these values micro- or macro-averages, or something else completely?
Thanks!
Using Spacy is it possible to get a past tense of a lemma word?
Example: take -> took
Hi guys, I want to avoid overfitting in my training custom NER model. Is feeding with 'enough' data the only solution? Is there another solution that you would recommend? How important is feeding data without training entities? E.g. from SpaCy documentation ("Do they bite?", {"entities": []}). Thanks a lot.
