feat: added trainable biaffine dependency parser and metrics

changelog.md

@@ -10,6 +10,7 @@
 - Support a specific model output dir path for trainings (`output_model_dir`), and whether to save the model or not (`save_model`)
 - Specify whether to log the validation results or not (`logger=False`)
 - Added support for the CoNLL format with `edsnlp.data.read_conll` and with a specific `eds.conll_dict2doc` converter
+- Added a Trainable Biaffine Dependency Parser (`eds.biaffine_dep_parser`) component and metrics
 
 ### Fixed
 

docs/pipes/trainable/biaffine-dependency-parser.md

@@ -0,0 +1,8 @@
+# Trainable Biaffine Dependency Parser {: #edsnlp.pipes.trainable.biaffine_dep_parser.factory.create_component }
+
+::: edsnlp.pipes.trainable.biaffine_dep_parser.factory.create_component
+    options:
+        heading_level: 2
+        show_bases: false
+        show_source: false
+        only_class_level: true

docs/pipes/trainable/index.md

@@ -8,13 +8,14 @@ All trainable components implement the [`TorchComponent`][edsnlp.core.torch_comp
 
 <!-- --8<-- [start:components] -->
 
-| Name                  | Description                                                           |
-|-----------------------|-----------------------------------------------------------------------|
-| `eds.transformer`     | Embed text with a transformer model                                   |
-| `eds.text_cnn`        | Contextualize embeddings with a CNN                                   |
-| `eds.span_pooler`     | A span embedding component that aggregates word embeddings            |
-| `eds.ner_crf`         | A trainable component to extract entities                             |
-| `eds.span_classifier` | A trainable component for multi-class multi-label span classification |
-| `eds.span_linker`     | A trainable entity linker (i.e. to a list of concepts)                |
+| Name                      | Description                                                           |
+|---------------------------|-----------------------------------------------------------------------|
+| `eds.transformer`         | Embed text with a transformer model                                   |
+| `eds.text_cnn`            | Contextualize embeddings with a CNN                                   |
+| `eds.span_pooler`         | A span embedding component that aggregates word embeddings            |
+| `eds.ner_crf`             | A trainable component to extract entities                             |
+| `eds.span_classifier`     | A trainable component for multi-class multi-label span classification |
+| `eds.span_linker`         | A trainable entity linker (i.e. to a list of concepts)                |
+| `eds.biaffine_dep_parser` | A trainable biaffine dependency parser                                |
 
 <!-- --8<-- [end:components] -->

docs/references.bib

@@ -161,3 +161,26 @@ @article{petitjean_2024
     url = {https://doi.org/10.1093/jamia/ocae069},
     eprint = {https://academic.oup.com/jamia/article-pdf/31/6/1280/57769016/ocae069.pdf},
 }
+
+@misc{dozat2017deepbiaffineattentionneural,
+      title={Deep Biaffine Attention for Neural Dependency Parsing},
+      author={Timothy Dozat and Christopher D. Manning},
+      year={2017},
+      eprint={1611.01734},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL},
+      url={https://arxiv.org/abs/1611.01734},
+}
+
+@inproceedings{grobol:hal-03223424,
+    title = {{Analyse en dépendances du français avec des plongements contextualisés}},
+    author = {Grobol, Loïc and Crabbé, Benoît},
+    url = {https://hal.archives-ouvertes.fr/hal-03223424},
+    year = {2021},
+    booktitle = {{Actes de la 28ème Conférence sur le Traitement Automatique des Langues Naturelles}},
+    eventtitle = {{TALN-RÉCITAL 2021}},
+    venue = {Lille, France},
+    pdf = {https://hal.archives-ouvertes.fr/hal-03223424/file/HOPS_final.pdf},
+    hal_id = {hal-03223424},
+    hal_version = {v1},
+}

edsnlp/metrics/dep_parsing.py

@@ -0,0 +1,55 @@
+from typing import Any, Optional
+
+from edsnlp import registry
+from edsnlp.metrics import Examples, make_examples, prf
+
+
+def dependency_parsing_metric(
+    examples: Examples,
+    filter_expr: Optional[str] = None,
+):
+    """
+    Compute the UAS and LAS scores for dependency parsing.
+
+    Parameters
+    ----------
+    examples : Examples
+        The examples to score, either a tuple of (golds, preds) or a list of
+        spacy.training.Example objects
+    filter_expr : Optional[str]
+        The filter expression to use to filter the documents
+
+    Returns
+    -------
+    Dict[str, float]
+    """
+    items = {
+        "uas": (set(), set()),
+        "las": (set(), set()),
+    }
+    examples = make_examples(examples)
+    if filter_expr is not None:
+        filter_fn = eval(f"lambda doc: {filter_expr}")
+        examples = [eg for eg in examples if filter_fn(eg.reference)]
+
+    for eg_idx, eg in enumerate(examples):
+        for token in eg.reference:
+            items["uas"][0].add((eg_idx, token.i, token.head.i))
+            items["las"][0].add((eg_idx, token.i, token.head.i, token.dep_))
+
+        for token in eg.predicted:
+            items["uas"][1].add((eg_idx, token.i, token.head.i))
+            items["las"][1].add((eg_idx, token.i, token.head.i, token.dep_))
+
+    return {name: prf(pred, gold)["f"] for name, (pred, gold) in items.items()}
+
+
+@registry.metrics.register("eds.dep_parsing")
+class DependencyParsingMetric:
+    def __init__(self, filter_expr: Optional[str] = None):
+        self.filter_expr = filter_expr
+
+    __init__.__doc__ = dependency_parsing_metric.__doc__
+
+    def __call__(self, *examples: Any):
+        return dependency_parsing_metric(examples, self.filter_expr)

edsnlp/pipes/__init__.py

@@ -75,6 +75,7 @@
     from .qualifiers.reported_speech.factory import create_component as reported_speech
     from .qualifiers.reported_speech.factory import create_component as rspeech
     from .trainable.ner_crf.factory import create_component as ner_crf
+    from .trainable.biaffine_dep_parser.factory import create_component as biaffine_dep_parser
     from .trainable.span_classifier.factory import create_component as span_classifier
     from .trainable.span_linker.factory import create_component as span_linker
     from .trainable.embeddings.span_pooler.factory import create_component as span_pooler

edsnlp/pipes/trainable/biaffine_dep_parser/__init__.py

@@ -0,0 +1 @@
+from .factory import create_component