uploading spacy model for ner.make-gold annotation

, , ,
1

Dear @ines and @honnibal,

I saw that some people had used the trained spacy model recursively for improving the accuracy of the NER tagging.

I trained an entity that is "SPORT" by using the spacy's train_new_entity_type.py with an existing "en_core_web_lg" model, and saved the output model for re-using it in the ner.make-gold. But when I run this command:
prodigy ner.make-gold sport_terms_2 /../prodigy/spaCy/examples/training/model_SPORT/ner/model ../raw_data_8_8_2019.jsonl --label "SPORT, ORG, GPE”
it did not start the API for annotation. May be I try to import wrong output of the spacy.

Could you please tell me how can I import the spacy_model that I obtained in order to use it for making gold data again?

Thanks,

2

What do you mean by "did not start the API for annotation"? Did you get an error?

The command you run looks correct, but this potentially wrong:

You'll need to load the entire model directory you exported when you pre-trained the model. The same thing that spaCy expects for spacy.load. It should be the directory that has a meta.json in it, and a bunch of directories for the pipeline components. From your example, it looks like you might be trying to just load the ner subdirectory?

3

I didn't get any error but I was waiting for a link to access the localhost:8080.
Actually I only tried to import the model file inside the ner subdirectory.

So, I tried to import the output directory that I got from the spacy training as you said, but there is no error it just give me ">" command line. I don't understand.

$ prodigy ner.make-gold sport_terms_2 /Users/asli/Desktop/NER/prodigy/spaCy/examples/training/model_SPORT ../raw_data_8_8_2019.jsonl --label "SPORT, ORG, GPE”

I want to use that model for annotating the new texts and make new annotations by using suggestions from the model so that I could use the new annotations for training spacy's train_new_entity_type.py again.
But, I guess ner.make-gold is not a method that suggests the entity from a given text, am I right? I would like to speed up annotation process for me since GPE and ORG entities are already in the pretrained model. So what do you suggest to me ?

I hope I am clear cause I am new to prodigy, and I get some difficulty to explain my problem with the syntactic language of the prodigy and spacy.

4

When I tried that command:

import spacy
spacy.load("/Users/asli/Desktop/NER/prodigy/spaCy/examples/training/model_SPORT")
<spacy.lang.en.English object at 0x10cb64320>
It seems like working but I cannot upload this model with:
prodigy ner.make-gold sport_terms_2 "/Users/asli/Desktop/NER/prodigy/spaCy/examples/training/model_SPORT" ../raw_data_8_8_2019.jsonl --label "SPORT, ORG, GPE”

I also checked the spaCy's Saving, loading and distributing models but I didn't understand clearly. Could you please help me.

5

I found my error. It's because of different type of double quote at the end of the labels. So it works now.

1 Like