Custom NER training and debug config.cfg file

[akki248]

Hi,

I'm trying to train a custom NER model.

Some useful details:
Python implementation: CPython
Python version : 3.8.8
IPython version : 7.22.0
Compiler : Clang 10.0.0
OS : Darwin
Release : 20.3.0
Machine : x86_64
Processor : i386
CPU cores : 8
Architecture: 64bit
spacy version: 3.0.6

I obtained the config.cfg file after running the command:

python -m spacy init fill-config base_config.cfg config.cfg

Then, I make no changes to the generated config.cfg file and run the following command as mentioned here for [debug].(https://spacy.io/api/cli#debug)

python -m spacy debug model config.cfg ner

But, I get the following error as shown in the attached snapshot:

If I ignore the error and train the custom NER model using the command:
python -m spacy train config.cfg --output ./output --paths.train ta_train.spacy --paths.dev ta_validation.spacy

It starts to train (regarding results, I'll start a new discussion).

I'm unable to understand that if there is an error regarding the tok2vec listener, how am I able to train and how does the output of the tok2vec layer go to the NER transition based parser?

Please let me know if you need any further information from my side.

Thanks.

[svlandeg]

Hi!

It looks like the generated config file doesn't have valid paths.train and paths.dev sections, but the debug model script needs some example data to initialize the pipeline.

Could you retry your debug command like this?

python -m spacy debug model config.cfg ner --paths.train ta_train.spacy --paths.dev ta_validation.spacy

[svlandeg]

Hi @akki248 : please don't artificially bump your thread as this is really unhelpful to us. We try to get to everyone's questions in due time, but with no garantuees.

[svlandeg]

I tried replicating your error with the config file you pasted. It refers to @Tokenizers which should be @tokenizers, so I'm guessing that you won't have been able to run this particular config file at all. You may have run another one.

Anyway, when I change L19 to

tokenizer = {"@tokenizers":"spacy.Tokenizer.v1"}

it runs just fine for me:

$ spacy debug model spacy/disc8268.cfg ner
ℹ Using CPU
ℹ To switch to GPU 0, use the option: --gpu-id 0
ℹ Fixing random seed: 0
ℹ Analysing model with ID 60
ℹ Initialized the model with dummy data.
✔ Succesfully ended analysis - model looks good.

[akki248]

Thanks for the correction. Even after changing the L19 to
tokenizer = {"@tokenizers":"spacy.Tokenizer.v1"}

I get the same error when I run the command to debug the ner. It's weird. But, when I run:
python -m spacy debug model config.cfg tok2vec

It seems to work.

Some useful details:
Python implementation: CPython
Python version : 3.8.8
IPython version : 7.22.0
Compiler : Clang 10.0.0
OS : Darwin
Release : 20.3.0
Machine : x86_64
Processor : i386
CPU cores : 8
Architecture: 64bit
spacy version: 3.0.6

Do you think there could be something else that I could check?

[svlandeg]

I'm afraid I can't be of any further assistance if I don't know which exact config file you've run. I suspect you have multiple config files and you may have been citing the wrong one in this thread, because it's not possible that the one with @Tokenizers in it, gave you the error as you said (nor is it possible that the error stayed the same after you changed the typo). The one you cited, worked fine for me (after the typo correction)

Also, as mentioned before, please do not share screenshots of code, but copy and paste the code as text instead (and format it properly with markdown), thanks.

[akki248]

OK. Thanks.