fix: Reuploaded previously unavailable SNL datasets

mteb/tasks/Clustering/nob/SNLHierarchicalClustering.py

@@ -18,11 +18,11 @@ class SNLHierarchicalClusteringP2P(AbsTaskClusteringFast):
     metadata = TaskMetadata(
         name="SNLHierarchicalClusteringP2P",
         dataset={
-            "path": "navjordj/SNL_summarization",
-            "revision": "3d3d27aa7af8941408cefc3991ada5d12a4273d1",
+            "path": "mteb/SNLHierarchicalClusteringP2P",
+            "revision": "693a321c42fb13ffe76bb9043f8d2aaa8f0a9499",
         },
         description="Webscrabed articles from the Norwegian lexicon 'Det Store Norske Leksikon'. Uses articles categories as clusters.",
-        reference="https://huggingface.co/datasets/navjordj/SNL_summarization",
+        reference="https://huggingface.co/datasets/mteb/SNLHierarchicalClusteringP2P",
         type="Clustering",
         category="p2p",
         modalities=["text"],
@@ -48,12 +48,6 @@ class SNLHierarchicalClusteringP2P(AbsTaskClusteringFast):
     )
     max_depth = 5
 
-    def dataset_transform(self) -> None:
-        self.dataset = self.dataset.rename_columns(
-            {"article": "sentences", "category": "labels"}
-        )
-        self.dataset = self.dataset.map(split_labels)
-
 
 class SNLHierarchicalClusteringS2S(AbsTaskClusteringFast):
     max_document_to_embed = 1300
@@ -62,11 +56,11 @@ class SNLHierarchicalClusteringS2S(AbsTaskClusteringFast):
     metadata = TaskMetadata(
         name="SNLHierarchicalClusteringS2S",
         dataset={
-            "path": "navjordj/SNL_summarization",
-            "revision": "3d3d27aa7af8941408cefc3991ada5d12a4273d1",
+            "path": "mteb/SNLHierarchicalClusteringS2S",
+            "revision": "b505e4ce65f255228e49dd07b6f8148731c5dc64",
         },
         description="Webscrabed articles from the Norwegian lexicon 'Det Store Norske Leksikon'. Uses articles categories as clusters.",
-        reference="https://huggingface.co/datasets/navjordj/SNL_summarization",
+        reference="https://huggingface.co/datasets/mteb/SNLHierarchicalClusteringS2S",
         type="Clustering",
         category="s2s",
         modalities=["text"],
@@ -91,9 +85,3 @@ class SNLHierarchicalClusteringS2S(AbsTaskClusteringFast):
         prompt="Identify categories in a Norwegian lexicon",
     )
     max_depth = 5
-
-    def dataset_transform(self) -> None:
-        self.dataset = self.dataset.rename_columns(
-            {"ingress": "sentences", "category": "labels"}
-        )
-        self.dataset = self.dataset.map(split_labels)

mteb/tasks/Clustering/nob/snl_clustering.py

@@ -1,12 +1,9 @@
 from __future__ import annotations
 
-import random
 from collections.abc import Iterable
 from itertools import islice
 from typing import TypeVar
 
-import datasets
-
 from mteb.abstasks.AbsTaskClustering import AbsTaskClustering
 from mteb.abstasks.TaskMetadata import TaskMetadata
 
@@ -27,11 +24,11 @@ class SNLClustering(AbsTaskClustering):
     metadata = TaskMetadata(
         name="SNLClustering",
         dataset={
-            "path": "navjordj/SNL_summarization",
-            "revision": "3d3d27aa7af8941408cefc3991ada5d12a4273d1",
+            "path": "mteb/SNLClustering",
+            "revision": "e1c801d5a6fe26c89d5e878181246f5b292e6549",
         },
         description="Webscrabed articles from the Norwegian lexicon 'Det Store Norske Leksikon'. Uses articles categories as clusters.",
-        reference="https://huggingface.co/datasets/navjordj/SNL_summarization",
+        reference="https://huggingface.co/datasets/mteb/SNLClustering",
         type="Clustering",
         category="p2p",
         modalities=["text"],
@@ -54,49 +51,3 @@ class SNLClustering(AbsTaskClustering):
 }
 """,
     )
-
-    def dataset_transform(self):
-        splits = self.metadata_dict["eval_splits"]
-
-        documents: list = []
-        labels: list = []
-        label_col = "category"
-
-        ds = {}
-        for split in splits:
-            ds_split = self.dataset[split]
-
-            _label = self.normalize_labels(ds_split[label_col])
-            documents.extend(ds_split["ingress"])
-            labels.extend(_label)
-
-            documents.extend(ds_split["article"])
-            labels.extend(_label)
-
-            assert len(documents) == len(labels)
-
-            rng = random.Random(42)  # local only seed
-            pairs = list(zip(documents, labels))
-            rng.shuffle(pairs)
-            documents, labels = (list(collection) for collection in zip(*pairs))
-
-            # reduce size of dataset to not have too large datasets in the clustering task
-            documents_batched = list(batched(documents, 512))[:4]
-            labels_batched = list(batched(labels, 512))[:4]
-
-            ds[split] = datasets.Dataset.from_dict(
-                {
-                    "sentences": documents_batched,
-                    "labels": labels_batched,
-                }
-            )
-
-        self.dataset = datasets.DatasetDict(ds)
-
-    @staticmethod
-    def normalize_labels(labels: list[str]) -> list[str]:
-        # example label:
-        # Store norske leksikon,Kunst og estetikk,Musikk,Klassisk musikk,Internasjonale dirigenter
-        # When using 2 levels there is 17 unique labels
-        # When using 3 levels there is 121 unique labels
-        return [",".join(tuple(label.split(",")[:3])) for label in labels]

mteb/tasks/Retrieval/nob/snl_retrieval.py

@@ -10,11 +10,11 @@ class SNLRetrieval(AbsTaskRetrieval):
     metadata = TaskMetadata(
         name="SNLRetrieval",
         dataset={
-            "path": "navjordj/SNL_summarization",
+            "path": "adrlau/navjordj-SNL_summarization_copy",  # TODO: replace with mteb/SNLRetrieval after #2820 is resolved.
             "revision": "3d3d27aa7af8941408cefc3991ada5d12a4273d1",
         },
         description="Webscrabed articles and ingresses from the Norwegian lexicon 'Det Store Norske Leksikon'.",
-        reference="https://huggingface.co/datasets/navjordj/SNL_summarization",
+        reference="https://huggingface.co/datasets/mteb/SNLRetrieval",
         type="Retrieval",
         category="p2p",
         modalities=["text"],

tests/test_tasks/test_all_abstasks.py

@@ -34,10 +34,6 @@
 
 datasets_not_available = [
     "AfriSentiLangClassification",
-    "SNLHierarchicalClusteringP2P",
-    "SNLClustering",
-    "SNLHierarchicalClusteringS2S",
-    "SNLRetrieval",
 ]