Access NER Tags in a converted .spacy file

[fariiaakh]

Hi Everyone,

Raphael and Thomas, who are a few of the SPACY developers helped me to convert my IOB file to a .spacy file.

However, when I asked about how I can view the NER tags in my .spacy converted file, I was told to use the '.label_' property.
#11391 (reply in thread)

The .label_ property works for accessing entities using pretrained spacy models. I want to train a model with my converted data but I want to see if the data converted properly first but I am unable to see that:

I know the '.label_ ' property works for doc objects, based on the spacy course, but what i have is a DocBin object and there is no property for extracting the entity label.

I converted and IOB file to a .spacy file.

This is what the iob file looks like:

When i converted my IOB file to json, i was able to see what the data looks like:

But i cannot see what my converted spacy file looks like. I can access the vocab with nlp.vocab but not the ner tags.

Please help!

[richardpaulhudson]

• Please don't paste images of code or terminal output; paste text as text. It's hard for us to read images and we can't modify your code or inspect, and we can't search it on the forums. You could have a look at this link that offers general guidelines about submitting issues and discussions on Github.

• I think what is perhaps a bit confusing here is that there are two views of the same entity data. You can set and access entity labels either at the Doc level e.g. using set_ents() or at the level of the individual tokens that make up a Doc. For example, token = doc[2] returns you the third token in the document. You can then access any entity information relating to that token via token.ent_type_ and token.ent_iob_: this is just a different view of the same information you set using set_ents(). See this link for more information.

• When you create a DocBin object, you can specify a list of token attributes to include for each document. The attribute labels for the information you are interested in here would be ENT_TYPE and ENT_IOB, so you could create a DocBin with doc_bin = DocBin(attrs=["ENT_IOB", "ENT_TYPE"]). However, you can also leave the attrs parameter out altogether, in which case a DocBin is created that stores a number of standard parameters that includes these two. See this link.

• When you retrieve the documents from a DocBin, you always specify the Vocab object from the pipeline with which the DocBin was created (see the example here). Once you've done that and got each document, you can access the entity information within a document using either the document-based or the token-based view.

[fariiaakh]

Thank You So Much Richard!! I've been stuck on this for weeks!! Thank You!!