Error spancat singlelabel config

[dad766]

Hello,

I am trying to use the singlelabel spancat defined here:
https://spacy.io/api/spancategorizer

I did not find a command in Prodigy to generate the cfg file directly so I had to do it manually.
my config file :

[paths]
train = tmp/train_for_spacy/train.spacy
dev = tmp/train_for_spacy/dev.spacy
vectors = null
init_tok2vec = null

[system]
seed = 0
gpu_allocator = null

[nlp]
lang = "fr"
pipeline = ["spancat_singlelabel"]
disabled = []
before_creation = null
after_creation = null
after_pipeline_creation = null
batch_size = 64
tokenizer = {"@tokenizers":"spacy.Tokenizer.v1"}

[components]

[components.spancat_singlelabel]
factory = "spancat_singlelabel"
allow_overlap = true
negative_weight = 0.8
scorer = {"@scorers":"spacy.spancat_scorer.v1"}
spans_key = "labeled_spans"

[components.spancat_singlelabel.model]
@architectures = "spacy.SpanCategorizer.v1"

[components.spancat_singlelabel.model.reducer]
@layers = "spacy.mean_max_reducer.v1"
hidden_size = 128

[components.spancat_singlelabel.model.scorer]
@layers = "Softmax.v2"
nO = null
nI = null
init_W = null
init_b = null
normalize_outputs = true
temperature = 1.0

[components.spancat_singlelabel.model.tok2vec]
@architectures = "spacy.Tok2Vec.v2"

[components.spancat_singlelabel.model.tok2vec.embed]
@architectures = "spacy.MultiHashEmbed.v1"
width = 96
rows = [5000,1000,2500,1000]
attrs = ["NORM","PREFIX","SUFFIX","SHAPE"]
include_static_vectors = false

[components.spancat_singlelabel.model.tok2vec.encode]
@architectures = "spacy.MaxoutWindowEncoder.v2"
width = 96
window_size = 1
maxout_pieces = 3
depth = 4

[components.spancat_singlelabel.suggester]
@misc = "spacy.ngram_range_suggester.v1"
min_size = 1
max_size = 62

[corpora]

[corpora.dev]
@readers = "spacy.Corpus.v1"
path = ${paths.dev}
gold_preproc = false
max_length = 0
limit = 0
augmenter = null

[corpora.train]
@readers = "spacy.Corpus.v1"
path = ${paths.train}
gold_preproc = false
max_length = 0
limit = 0
augmenter = null

[training]
seed = ${system.seed}
gpu_allocator = ${system.gpu_allocator}
dropout = 0.1
accumulate_gradient = 1
patience = 1600
max_epochs = 0
max_steps = 20000
eval_frequency = 200
frozen_components = []
annotating_components = []
dev_corpus = "corpora.dev"
train_corpus = "corpora.train"
before_to_disk = null
before_update = null
logger = {"@loggers":"spacy.ConsoleLogger.v1"}

[training.batcher]
@batchers = "spacy.batch_by_words.v1"
discard_oversize = false
tolerance = 0.2

[training.batcher.size]
@schedules = "compounding.v1"
start = 100
stop = 1000
compound = 1.001

[training.optimizer]
@optimizers = "Adam.v1"
beta1 = 0.9
beta2 = 0.999
L2_is_weight_decay = true
L2 = 0.01
grad_clip = 1.0
use_averages = false
eps = 0.00000001
learn_rate = 0.001

[training.score_weights]
spans_sc_f = 1.0
spans_sc_p = 0.0
spans_sc_r = 0.0

[initialize]
vectors = ${paths.vectors}
init_tok2vec = ${paths.init_tok2vec}
vocab_data = null
lookups = null
before_init = null
after_init = null

[initialize.components]

[initialize.tokenizer]

but when I try to do a training I get the message :

ValueError: [E143] Labels for component 'spancat_singlelabel' not initialized. This can be fixed by calling add_label, or by providing a representative batch of examples to the component's initialize method.

I have the impression that it is missing the label used. In the case of a classic spancat it is defined I think in the label/spancat.json directory. I tried that but it didn't work.
What should I do?

[dad766]

I try to create my config file like that to use add_label function. Not sure that's the best way...

from spacy.pipeline.spancat import DEFAULT_SPANCAT_SINGLELABEL_MODEL
import spacy

nlp = spacy.load("blank:fr")

config = {
    "spans_key": "labeled_spans",
    "model": DEFAULT_SPANCAT_SINGLELABEL_MODEL,
    "suggester": {"@misc": "spacy.ngram_suggester.v1", "sizes": [1, 2, 3]},
    # Additional spancat_singlelabel parameters
    "negative_weight": 0.8,
    "allow_overlap": False,
}
spancat = nlp.add_pipe("spancat_singlelabel", config=config)
spancat.add_label("Intentions exprim\u00e9es")


spancat.to_disk("tmp/pipeline")
nlp.to_disk("tmp/pipeline")

[kadarakos]

Hey there,

Sorry for the late response! Just ran an experiment on spancat_singlelabel to make sure it works properly.
You can generate a default spancat_singlelabel config with spacy:

python -m spacy init config configs/spancat_singlelabel.cfg --lang en --pipeline spancat_singlelabel

This example generates a config in configs/spancat_singlelabel.cfg with a pipeline for the language English.

In terms of the error you get, the first thing that comes to mind is to make sure that the spans_key in the config corresponds to the doc.spans["key"] where the spans are stored that you would like to work with.