add requirements to doc build

docs/requirements.txt

@@ -5,4 +5,9 @@ sphinx
 importlib-metadata
 sphinx-multiversion
 pydata-sphinx-theme<0.14
-sphinx_design
+sphinx_design
+
+# previous dependencies that are required to build docs for later versions too.
+semver
+gensim
+bpemb

scripts/span_tagger.py

@@ -0,0 +1,22 @@
+from flair.data import Sentence
+from flair.models import MultitaskModel
+
+# For comparison: This works since the label type is "ner" for both models in the multitask model
+classifier: MultitaskModel = MultitaskModel.load("zelda")
+
+sentence = Sentence("Kirk and Spock met on the Enterprise")
+
+classifier.predict(sentence)
+
+print(sentence)
+
+# Giving them sensible label names, now made possible with this PR
+classifier.tasks["Task_1"]._label_type = "nel"
+classifier.tasks["Task_1"]._span_label_type = "ner"
+
+# However, this no longer makes predictions
+sentence = Sentence("Kirk and Spock met on the Enterprise")
+
+classifier.predict(sentence)
+
+print(sentence)

scripts/train_span_tagger_multi.py

@@ -0,0 +1,58 @@
+from flair.datasets import NER_MULTI_WIKINER, ZELDA
+from flair.embeddings import TransformerWordEmbeddings
+from flair.models import SequenceTagger, SpanClassifier
+from flair.models.entity_linker_model import CandidateGenerator
+from flair.trainers import ModelTrainer
+from flair.nn import PrototypicalDecoder
+from flair.nn.multitask import make_multitask_model_and_corpus
+
+# 1. get the corpus
+ner_corpus = NER_MULTI_WIKINER().downsample(0.001)
+nel_corpus = (
+    ZELDA(column_format={0: "text", 2: "nel"})
+    .downsample(0.0001, downsample_dev=False, downsample_test=False)
+    .downsample(0.01, downsample_train=False)
+)  # need to set the label type to be the same as the ner one
+
+# --- Embeddings that are shared by both models --- #
+shared_embeddings = TransformerWordEmbeddings("distilbert-base-uncased", fine_tune=True)
+
+ner_label_dict = ner_corpus.make_label_dictionary("ner", add_unk=False)
+
+ner_model = SequenceTagger(
+    embeddings=shared_embeddings,
+    tag_dictionary=ner_label_dict,
+    tag_type="ner",
+    use_rnn=False,
+    use_crf=False,
+    reproject_embeddings=False,
+)
+
+
+nel_label_dict = nel_corpus.make_label_dictionary("nel", add_unk=True)
+
+nel_model = SpanClassifier(
+    embeddings=shared_embeddings,
+    label_dictionary=nel_label_dict,
+    label_type="nel",
+    span_label_type="ner",
+    decoder=PrototypicalDecoder(
+        num_prototypes=len(nel_label_dict),
+        embeddings_size=shared_embeddings.embedding_length * 2,  # we use "first_last" encoding for spans
+        distance_function="dot_product",
+    ),
+    candidates=CandidateGenerator("zelda"),
+)
+
+
+# -- Define mapping (which tagger should train on which model) -- #
+multitask_model, multicorpus = make_multitask_model_and_corpus(
+    [
+        (ner_model, ner_corpus),
+        (nel_model, nel_corpus),
+    ]
+)
+
+# -- Create model trainer and train -- #
+trainer = ModelTrainer(multitask_model, multicorpus)
+trainer.fine_tune(f"resources/taggers/zelda_with_mention", mini_batch_chunk_size=1)