Experimental CoreferenceResolver fails (KeyError: "Parameter 'E' for model 'hashembed' has not been allocated yet.")

[Jonathan-Buck]

Hi, I was interested in seeing if the experimental CoreferenceResolver would work for a problem I'm trying to solve in one of my projects. I decided to start by copying the examples from the documentation page and encountered this error. Any insight is appriciated.

How to reproduce the behaviour

import spacy
nlp = spacy.load("en_core_web_sm")

doc = nlp("I eat chicken with rice.")
coref = nlp.add_pipe("experimental_coref")

# This usually happens under the hood
processed = coref(doc)

This snippet was found on the coref documentation page

Stacktrace:

Traceback (most recent call last):
  File "/Users/jon/Documents/Projects/mitre-parsing/coref.py", line 8, in <module>
    processed = coref(doc)
  File "spacy/pipeline/trainable_pipe.pyx", line 56, in spacy.pipeline.trainable_pipe.TrainablePipe.__call__
  File "/opt/homebrew/lib/python3.10/site-packages/spacy/util.py", line 1630, in raise_error
    raise e
  File "spacy/pipeline/trainable_pipe.pyx", line 52, in spacy.pipeline.trainable_pipe.TrainablePipe.__call__
  File "/opt/homebrew/lib/python3.10/site-packages/spacy_experimental/coref/coref_component.py", line 153, in predict
    scores, idxs = self.model.predict([doc])
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/model.py", line 315, in predict
    return self._func(self, X, is_train=False)[0]
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/layers/chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/layers/chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/layers/chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/layers/with_array.py", line 30, in forward
    return _ragged_forward(
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/layers/with_array.py", line 89, in _ragged_forward
    Y, get_dX = layer(Xr.dataXd, is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/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 "/opt/homebrew/lib/python3.10/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 "/opt/homebrew/lib/python3.10/site-packages/thinc/model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/layers/chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/model.py", line 291, in __call__
    return self._func(self, X, is_train=is_train)
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/layers/hashembed.py", line 61, in forward
    vectors = cast(Floats2d, model.get_param("E"))
  File "/opt/homebrew/lib/python3.10/site-packages/thinc/model.py", line 216, in get_param
    raise KeyError(
KeyError: "Parameter 'E' for model 'hashembed' has not been allocated yet."

Similar issues I found:
ner
morphologizer

Your Environment

Info about spaCy

• spaCy version: 3.3.1
• Platform: macOS-12.6-arm64-arm-64bit
• Python version: 3.10.8
• Pipelines: en_core_web_lg (3.3.0), en_core_web_sm (3.3.0)

[polm]

Sorry this isn't perfectly clear, but that part of the docs is designed to give you an idea of how things work, and not to be copy-pasted. In particular the last line is just intended to illustrate how the component is used and definitely cannot be just run after the preceding lines.

This particular error is because the component you just added hasn't been trained, so it's uninitialized. You need to initialize the component first, at which point you won't get an error, but you also won't get meaningful results until you train it.

For getting started with coref, I would recommend checking out the sample project. The API docs are more details if you need to tweak things or learn about the internals.

[Jonathan-Buck]

Gotcha, that makes sense. I guess the answers to the other questions i found were more helpful than I thought. I'll check out the sample project thanks for linking it!