spacy ver 3.2.0: Error when adding multiple ner and entity_ruler pipes to pipeline in code.

[smisra1]

I am switching from spacy ver 2 to 3.

I want my spacy pipeline in the following structure: ["entity_ruler_1", "ner_1", "entity_ruler_2", "ner_2", "entity_ruler_3", "ner_3"]

Below is a dummy code to get the error:

import spacy
#nlp = spacy.load('models/01')
nlp = spacy.blank('en')
er_1_list = ['car', 'product']
er_1_dict = [{'label': 'car', 'pattern': 'honda'}, {'label': 'car', 'pattern': 'hondas'}, {'label': 'car', 'pattern': 'hyundai'}]

er_2_list = ['food', 'type']
er_2_dict = [{'label': 'food', 'pattern': 'waffles'}, {'label': 'food', 'pattern': 'waffle'}, {'label': 'type', 'pattern': 'breakfasts'}]

er_3_list = ['forest', 'animals']
er_3_dict = [{'label': 'animals', 'pattern': 'dog'}, {'label': 'animals', 'pattern': 'puppy'}]

er1 = nlp.add_pipe("entity_ruler", name="entity_ruler_1")
er1.add_patterns(er_1_dict)

ner1 = nlp.add_pipe("ner", name="ner_1")
for i in er_1_list:
    ner1.add_label(i)

er2 = nlp.add_pipe("entity_ruler", name="entity_ruler_2")
er2.add_patterns(er_2_dict)

ner2 = nlp.add_pipe("ner", name="ner_2")
for i in er_2_list:
    ner1.add_label(i)
er3 = nlp.add_pipe("entity_ruler", name="entity_ruler_3")
er3.add_patterns(er_3_dict)

ner3 = nlp.add_pipe("ner", name="ner_3")
for i in er_3_list:
    ner1.add_label(i)

The error I get:

Traceback (most recent call last):
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_test.py", line 19, in <module>
    er2.add_patterns(er_2_dict)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\spacy\pipeline\entityruler.py", line 315, in add_patterns
    for label, pattern, ent_id in zip(
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\spacy\language.py", line 1578, in pipe
    for doc in docs:
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\spacy\util.py", line 1599, in _pipe
    yield from proc.pipe(docs, **kwargs)
  File "spacy\pipeline\transition_parser.pyx", line 240, in pipe
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\spacy\util.py", line 1618, in raise_error
    raise e
  File "spacy\pipeline\transition_parser.pyx", line 236, in spacy.pipeline.transition_parser.Parser.pipe
  File "spacy\pipeline\transition_parser.pyx", line 250, in spacy.pipeline.transition_parser.Parser.predict
  File "spacy\pipeline\transition_parser.pyx", line 265, in spacy.pipeline.transition_parser.Parser.greedy_parse
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 315, in predict
    return self._func(self, X, is_train=False)[0]
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\spacy\ml\tb_framework.py", line 33, in forward
    step_model = ParserStepModel(
  File "spacy\ml\parser_model.pyx", line 216, in spacy.ml.parser_model.ParserStepModel.__init__
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\with_array.py", line 30, in forward
    return _ragged_forward(
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\with_array.py", line 90, in _ragged_forward
    Y, get_dX = layer(Xr.dataXd, is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\concatenate.py", line 44, in forward
    Ys, callbacks = zip(*[layer(X, is_train=is_train) for layer in model.layers])
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\concatenate.py", line 44, in <listcomp>
    Ys, callbacks = zip(*[layer(X, is_train=is_train) for layer in model.layers])
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\layers\hashembed.py", line 61, in forward
    vectors = cast(Floats2d, model.get_param("E"))
  File "C:\Users\smisr\PycharmProjects\spacy_3_2_0\spacy_3\lib\site-packages\thinc\model.py", line 216, in get_param
    raise KeyError(
KeyError: "Parameter 'E' for model 'hashembed' has not been allocated yet."

I would like to stick to this way of model training only instead of using the terminal.
Please let me know what I need to change and why the error is occurring.

[polm]

I think that error is happening because you don't have a tok2vec - NER layers need vectorized features to work with, that can be provided by a tok2vec or a transformer.

Your model structure is a little different from what we usually see, so we don't have an exact example for it. What I would recommend you do is use the Quickstart to generate a pipeline with just one NER component and look at how it's structured, in particular the pipeline entry. If you can use that config (even with an in-code training loop) to train one of your NER models then you know your structure is OK and you can work on adding in your other NER models and entity rulers.