Finding hyperparameters of given model

[Jiya126]

How to find the hyperparameters in "en_coreference_web_trf" model
On using

nlp = spacy.load("en_coreference_web_trf")
 
doc = nlp(text)

# Print out component names
msg.info("Pipeline components")
for i, pipe in enumerate(nlp.pipe_names):
    print(f"{i}: {pipe}")

    print("###################################################")
    
    gpipe = nlp.get_pipe(pipe)
    hyperparameter = gpipe.get_config()
    print(hyperparameter)

It is throwing error:

ℹ Pipeline components
0: sentencizer
###################################################

AttributeError Traceback (most recent call last)
in <cell line: 17>()
21
22 gpipe = nlp.get_pipe(pipe)
---> 23 hyperparameter = gpipe.get_config()
24 print(hyperparameter)
25

AttributeError: 'Sentencizer' object has no attribute 'get_config'

And similarly for other components and models as well

Is there any way to access the hyperparameters of the model we are using?

Also, another thing, what is the difference between "en_coreference_web_trf" and "en_core_web_trf".
What I know is "en_coreference_web_trf" is in experimental branch of spacy and not mentioned in its documentation

[rmitsch]

Hey @Jiya126, you can access information about the components' config via nlp.config["components"].
en_coreference_web_trf includes a pretrained coref component - see here for more information.

[Jiya126]

Thank you, this works
But how can I change these hyperparameters, since it throws me an error stating config file cannot be overwritten

[Jiya126]

Also, where could I find documentation for en_coreference_web_trf
I'm unable to find that on https://spacy.io/

[svlandeg]

Hi, you can find more documentation in the following places:

• https://spacy.io/api/coref
• https://spacy.io/api/span-resolver
• https://spacy.io/api/architectures#Coref
• https://github.com/explosion/projects/tree/v3/experimental/coref