Default coreference usage runs into: "Did you forget to call initialize()? "

[mazieres]

Discussed in #13060

^{Originally posted by mazieres October 11, 2023}
I'm trying to use the default experimental coreference feature.

On a fresh python3 venv with spacy==3.6.0 and spacy-experimental, I followed the coref tutorial, as such:

nlp = spacy.load("en_core_web_sm")
coref = nlp.add_pipe("experimental_coref")
doc = nlp("This is a sentence.")

And ran into this error (full log below):

ValueError: [E109] Component 'experimental_coref' could not be run. Did you forget to call `initialize()`?

I've encountered many posts on the web with comparable errors but none of the solution fixed my problem. Any idea where/what I could look for ?

Thanks in advance for your help.

Full error log

File ~/.tmp/lib/python3.11/site-packages/spacy/language.py:1042, in Language.__call__(self, text, disable, component_cfg)
   1041 try:
-> 1042     doc = proc(doc, **component_cfg.get(name, {}))  # type: ignore[call-arg]
   1043 except KeyError as e:
   1044     # This typically happens if a component is not initialized

File ~/.tmp/lib/python3.11/site-packages/spacy/pipeline/trainable_pipe.pyx:56, in spacy.pipeline.trainable_pipe.TrainablePipe.__call__()

File ~/.tmp/lib/python3.11/site-packages/spacy/util.py:1724, in raise_error(proc_name, proc, docs, e)
   1723 def raise_error(proc_name, proc, docs, e):
-> 1724     raise e

File ~/.tmp/lib/python3.11/site-packages/spacy/pipeline/trainable_pipe.pyx:52, in spacy.pipeline.trainable_pipe.TrainablePipe.__call__()

File ~/.tmp/lib/python3.11/site-packages/spacy_experimental/coref/coref_component.py:153, in CoreferenceResolver.predict(self, docs)
    151     continue
--> 153 scores, idxs = self.model.predict([doc])
    154 # idxs is a list of mentions (start / end idxs)
    155 # each item in scores includes scores and a mapping from scores to mentions

File ~/.tmp/lib/python3.11/site-packages/thinc/model.py:334, in Model.predict(self, X)
    331 """Call the model's `forward` function with `is_train=False`, and return
    332 only the output, instead of the `(output, callback)` tuple.
    333 """
--> 334 return self._func(self, X, is_train=False)[0]

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/chain.py:54, in forward(model, X, is_train)
     53 for layer in model.layers:
---> 54     Y, inc_layer_grad = layer(X, is_train=is_train)
     55     callbacks.append(inc_layer_grad)

File ~/.tmp/lib/python3.11/site-packages/thinc/model.py:310, in Model.__call__(self, X, is_train)
    308 """Call the model's `forward` function, returning the output and a
    309 callback to compute the gradients via backpropagation."""
--> 310 return self._func(self, X, is_train=is_train)

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/chain.py:54, in forward(model, X, is_train)
     53 for layer in model.layers:
---> 54     Y, inc_layer_grad = layer(X, is_train=is_train)
     55     callbacks.append(inc_layer_grad)

File ~/.tmp/lib/python3.11/site-packages/thinc/model.py:310, in Model.__call__(self, X, is_train)
    308 """Call the model's `forward` function, returning the output and a
    309 callback to compute the gradients via backpropagation."""
--> 310 return self._func(self, X, is_train=is_train)

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/chain.py:54, in forward(model, X, is_train)
     53 for layer in model.layers:
---> 54     Y, inc_layer_grad = layer(X, is_train=is_train)
     55     callbacks.append(inc_layer_grad)

File ~/.tmp/lib/python3.11/site-packages/thinc/model.py:310, in Model.__call__(self, X, is_train)
    308 """Call the model's `forward` function, returning the output and a
    309 callback to compute the gradients via backpropagation."""
--> 310 return self._func(self, X, is_train=is_train)

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/with_array.py:36, in forward(model, Xseq, is_train)
     35 if isinstance(Xseq, Ragged):
---> 36     return cast(Tuple[SeqT, Callable], _ragged_forward(model, Xseq, is_train))
     37 elif isinstance(Xseq, Padded):

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/with_array.py:91, in _ragged_forward(model, Xr, is_train)
     90 layer: Model[ArrayXd, ArrayXd] = model.layers[0]
---> 91 Y, get_dX = layer(Xr.dataXd, is_train)
     93 def backprop(dYr: Ragged) -> Ragged:

File ~/.tmp/lib/python3.11/site-packages/thinc/model.py:310, in Model.__call__(self, X, is_train)
    308 """Call the model's `forward` function, returning the output and a
    309 callback to compute the gradients via backpropagation."""
--> 310 return self._func(self, X, is_train=is_train)

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/concatenate.py:57, in forward(model, X, is_train)
     56 def forward(model: Model[InT, OutT], X: InT, is_train: bool) -> Tuple[OutT, Callable]:
---> 57     Ys, callbacks = zip(*[layer(X, is_train=is_train) for layer in model.layers])
     58     if isinstance(Ys[0], list):

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/concatenate.py:57, in <listcomp>(.0)
     56 def forward(model: Model[InT, OutT], X: InT, is_train: bool) -> Tuple[OutT, Callable]:
---> 57     Ys, callbacks = zip(*[layer(X, is_train=is_train) for layer in model.layers])
     58     if isinstance(Ys[0], list):

File ~/.tmp/lib/python3.11/site-packages/thinc/model.py:310, in Model.__call__(self, X, is_train)
    308 """Call the model's `forward` function, returning the output and a
    309 callback to compute the gradients via backpropagation."""
--> 310 return self._func(self, X, is_train=is_train)

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/chain.py:54, in forward(model, X, is_train)
     53 for layer in model.layers:
---> 54     Y, inc_layer_grad = layer(X, is_train=is_train)
     55     callbacks.append(inc_layer_grad)

File ~/.tmp/lib/python3.11/site-packages/thinc/model.py:310, in Model.__call__(self, X, is_train)
    308 """Call the model's `forward` function, returning the output and a
    309 callback to compute the gradients via backpropagation."""
--> 310 return self._func(self, X, is_train=is_train)

File ~/.tmp/lib/python3.11/site-packages/thinc/layers/hashembed.py:62, in forward(model, ids, is_train)
     59 def forward(
     60     model: Model[Ints1d, OutT], ids: Ints1d, is_train: bool
     61 ) -> Tuple[OutT, Callable]:
---> 62     vectors = cast(Floats2d, model.get_param("E"))
     63     nV = vectors.shape[0]

File ~/.tmp/lib/python3.11/site-packages/thinc/model.py:235, in Model.get_param(self, name)
    234 if not self._params.has_param(self.id, name):
--> 235     raise KeyError(
    236         f"Parameter '{name}' for model '{self.name}' has not been allocated yet."
    237     )
    238 return self._params.get_param(self.id, name)

KeyError: "Parameter 'E' for model 'hashembed' has not been allocated yet."

The above exception was the direct cause of the following exception:

ValueError                                Traceback (most recent call last)
Cell In[6], line 1
----> 1 doc = nlp("This is a sentence.")

File ~/.tmp/lib/python3.11/site-packages/spacy/language.py:1045, in Language.__call__(self, text, disable, component_cfg)
   1042     doc = proc(doc, **component_cfg.get(name, {}))  # type: ignore[call-arg]
   1043 except KeyError as e:
   1044     # This typically happens if a component is not initialized
-> 1045     raise ValueError(Errors.E109.format(name=name)) from e
   1046 except Exception as e:
   1047     error_handler(name, proc, [doc], e)

ValueError: [E109] Component 'experimental_coref' could not be run. Did you forget to call `initialize()`?

Info about spaCy

• spaCy version: 3.6.0
• Platform: Linux-6.1.0-13-amd64-x86_64-with-glibc2.36
• Python version: 3.11.2
• Pipelines: en_core_web_sm (3.6.0), en_core_web_trf (3.6.1)

[rmitsch]

Hi @mazieres, please provide your code and config so we can look into this. Thanks!

[mazieres]

Hi @rmitsch, my post includes the packages installed, the code and its output together with the requested Info about SpaCy section. What else can I provide ?

[adrianeboyd]

Hi, you're trying to run an uninitialized/untrained new coreference component rather than loading an existing trained model. You can install our experimental demo English model following the directions here:

https://github.com/explosion/spacy-experimental/releases/tag/v0.6.1

[sahilmtayade]

How do we actually use this pre-trained model? I cannot find any example code or documentation for using the pre-trained model. Thanks in advance.

[sahilmtayade]

I found the answer, it was super hard to find. #11585

This shows how to use the default which is good enough for school projects like mine!

[mazieres]

This is the workaround I use but I was hoping to use the new default nlp.add_pipe("experimental_coref") which yields the error mentionned in the issue...

[adrianeboyd]

The demo model linked above does use the new experimental coref component, trained on English OntoNotes data.

If you use nlp.add_pipe("experimental_coref") you are adding a new component that hasn't been initialized or trained at all.

If you'd like to train your own coref model from scratch, this project is a good starting point: https://github.com/explosion/projects/tree/v3/experimental/coref

You can clone it with this command, but be aware you would need access to OntoNotes to run the project as-is:

spacy project clone experimental/coref