from datasets import Dataset, DatasetDict
import spacy
from spacy.tokens import DocBin
from pathlib import Path

nlp = spacy.blank("da")

corpus_path = Path("corpus/cdt_ddt")

def docbin_to_rows(path):
    doc_bin = DocBin().from_disk(path)
    rows = []
    for doc in doc_bin.get_docs(nlp.vocab):
        row = doc.to_json()
        row.pop("spans", None)  # dropped the key variable dictionaries to replace with a list representing the clusters

        row["coref_clusters"] = [
            [{"start": span.start_char, "end": span.end_char, "label": span.label_}
             for span in doc.spans[key]]
            for key in sorted(doc.spans.keys())
            if key.startswith("coref_clusters_")
        ]
        rows.append(row)
    return rows

dataset = DatasetDict({
    "train": Dataset.from_list(docbin_to_rows(corpus_path/"train.spacy")),
    "dev":   Dataset.from_list(docbin_to_rows(corpus_path/"dev.spacy")),
    "test":  Dataset.from_list(docbin_to_rows(corpus_path/"test.spacy")),
})

dataset.push_to_hub("johanamayer/cdt_ddt_union")