New Entity Tag replacing old Entities tag in additional entity type Training

[monk1337]

@honnibal I am trying to Train Spacy for custom Entity Label : My Entity is 'New_tag' and I am trying to train model with the previous model, so I want to append the training data with custom tag into existing model. I modified the default training script spacy/examples/training/train_new_entity_type.py a little my current code is

This is my code:

from __future__ import unicode_literals, print_function

import plac
import random
from pathlib import Path
import spacy

# new entity label
LABEL = 'New_tag'

TRAIN_DATA = [('How is Bitcoin doing today?', {'entities': [(15, 20, 'New_tag'), (21, 26, 'New_tag'), (7, 14, 'New_tag')]}), ('What People are talking about Bitcoin?', {'entities': [(5, 11, 'New_tag'), (16, 23, 'New_tag'), (30, 37, 'New_tag')]}), ('How Bitcoin will be doing tomorrow?', {'entities': [(4, 11, 'New_tag'), (26, 34, 'New_tag'), (20, 25, 'New_tag')]}), ('How Bitcoin will be doing one week?', {'entities': [(4, 11, 'New_tag'), (26, 34, 'New_tag'), (30, 34, 'New_tag')]}), ('How Bitcoin will be doing one month ahead?', {'entities': [(4, 11, 'New_tag'), (20, 25, 'New_tag'), (30, 35, 'New_tag'), (26, 35, 'New_tag'), (36, 41, 'New_tag')]}), ('What are the factors influencing Bitcoin prices?', {'entities': [(13, 20, 'New_tag'), (33, 40, 'New_tag'), (21, 32, 'New_tag')]})]

@plac.annotations(
    model=("Model name. Defaults to blank 'en' model.", "option", "m", str),
    new_model_name=("New model name for model meta.", "option", "nm", str),
    output_dir=("Optional output directory", "option", "o", Path),
    n_iter=("Number of training iterations", "option", "n", int))
def main(model='en', new_model_name='animal', output_dir='/Users/exepaul/', n_iter=20):
    """Set up the pipeline and entity recognizer, and train the new entity."""
    if model is not None:
        nlp = spacy.load(model)  # load existing spaCy model
        print("Loaded model '%s'" % model)

    nlp = spacy.load('en')

    # Add entity recognizer to model if it's not in the pipeline
    # nlp.create_pipe works for built-ins that are registered with spaCy
    if 'ner' not in nlp.pipe_names:
        ner = nlp.create_pipe('ner')
        nlp.add_pipe(ner)
    # otherwise, get it, so we can add labels to it
    else:
        ner = nlp.get_pipe('ner')

    ner.add_label(LABEL)   # add new entity label to entity recognizer

    # get names of other pipes to disable them during training
    other_pipes = [pipe for pipe in nlp.pipe_names if pipe != 'ner']
    with nlp.disable_pipes(*other_pipes):  # only train NER
        optimizer = nlp.begin_training()
        for itn in range(n_iter):
            random.shuffle(TRAIN_DATA)
            losses = {}
            for text, annotations in TRAIN_DATA:
                nlp.update([text], [annotations], sgd=optimizer, drop=0.35,
                           losses=losses)
            print(losses)

    # test the trained model
    test_text = 'What People are talking about Bitcoin?'
    doc = nlp(test_text)
    print("Entities in '%s'" % test_text)
    for ent in doc.ents:
        print(ent.label_, ent.text)

    # save model to output directory
    if output_dir is not None:
        output_dir = Path(output_dir)
        if not output_dir.exists():
            output_dir.mkdir()
        nlp.meta['name'] = new_model_name  # rename model
        nlp.to_disk(output_dir)
        print("Saved model to", output_dir)

        # test the saved model
        print("Loading from", output_dir)
        nlp2 = spacy.load(output_dir)
        doc2 = nlp2(test_text)
        for ent in doc2.ents:
            print(ent.label_, ent.text)

if __name__ == '__main__':
    plac.call(main)

Now If I load the model in another file and do this test :

import spacy

nlp=spacy.load('/Users/exepaul/google/models')

text=nlp(u'What People are talking about Bitcoin and Apple is looking at buying U.K. startup for $1 billion')
for i in text.ents:
    print(i.label_,i.text)

What i am getting is :

New_tag People New_tag talking New_tag Bitcoin New_tag Apple New_tag looking New_tag buying GPE U.K. MONEY $1 billion

But the expected result is:

New_tag People New_tag talking New_tag Bitcoin ORG Apple GPE U.K. MONEY $1 billion

how I can achieve this result and can add new entity type into existing model without replacing or modifying existing data or Entity types?

I gone through these issues but couldn't fi a d solution:

https://github.com/explosion/spaCy/issues/1752 https://github.com/explosion/spaCy/issues/1739 https://github.com/explosion/spaCy/issues/1585

[PhaneendraGunda]

Please don't include new entity types to the existing model until and unless if you have many examples. Iy you see, each entity in the spacy model have huge number of examples. So create a new model for your new entities and spacy model for default entities which works best when you have very small set of data.

[ines]

Copying over my reply from #2124:

It sounds like what you're experiencing is the "catastrophic forgetting" problem – as your model learns about the new entities, it's "forgetting" what it had previously learned.

Here are some resources and threads with more information and possible solutions:

• https://explosion.ai/blog/pseudo-rehearsal-catastrophic-forgetting

• 1892: Lot of false positives when using the NER model

• 1777: Improve spacy model for MONEY entity recognition

• 1337: Custom NER model doesn't recognize any entities

• 1382: Predefined entities not detected after adding custom entities

[lock[bot]]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.