Upload turl_table_col_type.py

Upload TURL Table Column Type Annotation Dataset Loading Script

turl_table_col_type.py

@@ -0,0 +1,150 @@
+"""This dataset is for the task of column type annotation"""
+import json
+import datasets
+
+_CITATION = """\
+@article{deng2020turl,
+  title={TURL: table understanding through representation learning},
+  author={Deng, Xiang and Sun, Huan and Lees, Alyssa and Wu, You and Yu, Cong},
+  journal={ACM SIGMOD Record},
+  volume={51},
+  number={1},
+  pages={33--40},
+  year={2022},
+  publisher={ACM New York, NY, USA}
+}"""
+
+_DESCRIPTION = """\
+Column Type Annotation(TURL)
+"""
+
+_HOMEPAGE = "https://github.com/sunlab-osu/TURL"
+
+_GIT_ARCHIVE_URL = "https://huggingface.co/datasets/stanford-crfm/helm-scenarios/tree/main/turl-column-type-annotation"
+
+_LICENSE = "CC BY 4.0"
+
+class ColumnAnnotation(datasets.GeneratorBasedBuilder):
+    """The TURL Column Annotation dataset"""
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "page_title" : datasets.Value("string"),
+                    "section_title" : datasets.Value("string"),
+                    "table_caption" : datasets.Value("string"),
+                    "table": {"header": datasets.features.Sequence(datasets.Value("string")),
+                              "rows": datasets.features.Sequence(datasets.features.Sequence(datasets.Value("string")))},
+                    "vocab": datasets.features.Sequence(datasets.Value("string")),
+                    "colname": datasets.Value("string"),
+                    "annotations": datasets.features.Sequence(datasets.Value("string")),
+                }
+            ),
+            supervised_keys=None,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+    
+    # To understand the json fields extracted here, refer to https://github.com/sunlab-osu/TURL?tab=readme-ov-file#column-type-annotation 
+    def _load_table(self, table_data):
+        headers = table_data[5]
+        col_count = len(table_data[6])
+        row_count = max([x[-1][0][0] for x in table_data[6]])  
+        rows = []
+        for i in range(row_count):
+            row = []
+            for j in range(col_count):
+                try:
+                    column_json = table_data[6][j]
+                    k = -1
+                    while k < row_count:
+                        k = k + 1
+                        if column_json[k][0][0] == i:
+                            break
+                    
+                    if k < row_count:
+                        val = table_data[6][j][k][1][1]
+                    else:
+                        val = ''
+                except:
+                    val = ''
+                row.append(val)
+            if not all(value == '' for value in row):
+                rows.append(row)
+        return {
+            "header": headers,
+            "rows": rows,
+        }
+    
+    def _split_generators(self, dl_manager):
+        """Returns SplitGenerators."""
+
+        from huggingface_hub import hf_hub_download
+
+        # Specify the repository and sub-directory
+        repo_id = "stanford-crfm/helm-scenarios"
+        sub_dir = "turl-column-type-annotation"
+
+        # Specify file names
+        train_file_name = "train.table_col_type.json"
+        dev_file_name = "dev.table_col_type.json"
+        test_file_name = "test.table_col_type.json"
+        vocab_file_name = "type_vocab.txt"
+
+        # Download files from hf
+        train_file_path = f"{sub_dir}/{train_file_name}"
+        train_file_path = hf_hub_download(repo_id, filename=train_file_path, repo_type="dataset", revision="main")
+
+        dev_file_path = f"{sub_dir}/{dev_file_name}"
+        dev_file_path = hf_hub_download(repo_id, filename=dev_file_path, repo_type="dataset", revision="main")
+
+        test_file_path = f"{sub_dir}/{test_file_name}"
+        test_file_path = hf_hub_download(repo_id, filename=test_file_path, repo_type="dataset", revision="main")
+
+        vocab_file_path = f"{sub_dir}/{vocab_file_name}"
+        vocab_file_path = hf_hub_download(repo_id, filename=vocab_file_path, repo_type="dataset", revision="main")
+
+        return [
+             datasets.SplitGenerator(
+                 name=datasets.Split.TRAIN,
+                 gen_kwargs={"vocabpath": vocab_file_path, "filepath": train_file_path},
+             ),
+             datasets.SplitGenerator(
+                 name=datasets.Split.VALIDATION,
+                 gen_kwargs={"vocabpath": vocab_file_path, "filepath": dev_file_path},
+             ),
+             datasets.SplitGenerator(
+                 name=datasets.Split.TEST,
+                 gen_kwargs={"vocabpath": vocab_file_path, "filepath": test_file_path},
+             ),
+        ]
+
+    def _generate_examples(self, vocabpath, filepath):
+        with open(filepath, 'r', encoding='utf-8') as json_file:
+            data = json.load(json_file)
+
+
+        with open(vocabpath, 'r', encoding='utf-8') as txt_file:
+            lines = txt_file.readlines()
+            vocab = [line[:-1].split('\t')[1] for line in lines]
+
+        index = 0    
+        for idx, table_data in enumerate(data):
+            # Load table contents using the _load_table method
+            table_content = self._load_table(table_data)
+            for col_idx in range(len(table_data[5])):
+                yield index, {
+                    "page_title" : table_data[1],
+                    "section_title" : table_data[3],
+                    "table_caption" : table_data[4],
+                    "table": table_content,
+                    "vocab": vocab,
+                    "colname": table_data[5][col_idx],
+                    "annotations": table_data[7][col_idx],
+                }
+                index += 1
+
+