Builder in one file

.gitignore

@@ -1,2 +1,4 @@
 env
-/env
+/env
+.vscode
+/.vscode

.vscode/settings.json

@@ -1,3 +0,0 @@
-{
-  "cSpell.words": ["Multiclassification"]
-}

py_legislation.py

@@ -14,13 +14,149 @@ class PY_Legislation(datasets.GeneratorBasedBuilder)
 Defines the implementation of Paraguay Legislation dataset builder (GeneratorBasedBuilder).
 
 """
+import textwrap
 
 from textwrap import TextWrapper
 import datasets
 import pyarrow.parquet as pq
 
-from features_specs import BASIC_FEATURES_SPEC, RAW_FEATURES_SPEC, SENTENCES_UNLABELED_FEATURES_SPEC
-from py_legislation_metadata import PY_LEGISLATION_METADATA
+from obligations import affected_entity, cost_type, aa_categories, aa_categories_unique, io_categories
+
+_metadata = {
+    "citation": """\
+    @InProceedings{
+        huggingface:dataset,
+        title = {Paraguay Legislation Dataset},
+        author={Peres, Fernando; Costa, Victor},
+        year={2023}
+    }
+    """,
+
+    "description": textwrap.dedent("""\
+        Dataset for researching - NLP techniques on PARAGUAY legislation.
+    """),
+
+    "homepage": "https://www.leyes.com.py/",
+
+    "license": "apache-2.0",
+
+    "urls": {
+        "raw": "./data/0_raw/raw.parquet",
+        "sentences_unlabeled": "./data/1_sentences_unlabeled/unlabeled.parquet",
+        "sentences_labeled": "./data/2_sentences_labeled/labeled.parquet",
+    },
+
+    # [@] Config Names:
+
+    "raw": {
+        "description": textwrap.dedent("""
+        Data extracted from the sources files (URls, PDFs and Word files) without any transformation or sentence splitter. It can be helpful because you can access the raw data extracted from the seeds (PDFs and Word files) and apply other preprocessing tasks from this point to prepare the data without returning to extract texts from source files.
+        """),
+
+        "features": {
+            "source_id": datasets.Value(dtype="int64"),
+            "doc_source_id": datasets.Value(dtype="int64"),
+            "document": datasets.Value(dtype="string"),
+            "text": datasets.Value(dtype="string"),
+        }
+
+    },
+
+    "sentences-unlabeled": {
+        "description": textwrap.dedent("""
+        Unlabeled corpora of Paraguay legislation. This data is prepared to be labeled by the experts. Each instance of the dataset represents a specific text passage, split by its original formatting extracted from raw text (from original documents)
+
+        Each observation of the dataset represents a specific text passage.
+        """),
+
+        "features": {
+            "source_id": datasets.Value(dtype="int64"),
+            "doc_source_id": datasets.Value(dtype="int64"),
+            "document": datasets.Value(dtype="string"),
+            "text": datasets.Value(dtype="string"),
+
+            # Categories
+            "cost_type": datasets.ClassLabel(names=cost_type,),
+            "affected_entity": datasets.ClassLabel(names=affected_entity,),
+            "io_categories": datasets.Sequence(
+                datasets.ClassLabel(names=io_categories,)),
+            "aa_categories": datasets.Sequence(
+                datasets.ClassLabel(names=aa_categories,)),
+            "aa_categories_unique": datasets.Sequence(
+                datasets.ClassLabel(names=aa_categories_unique,)),
+        }
+    },
+
+    "sentences-labeled": {
+        "description": textwrap.dedent("""
+        The labeled data is the ground truth data used to train the models. This data is annotated  by legal experts indicating the existence of administrative costs (and other types) in the legislation.
+
+        Each observation of the dataset represents a specific text passage.
+        """),
+
+        "features": {
+            "source_id": datasets.Value(dtype="int64"),
+            "doc_source_id": datasets.Value(dtype="int64"),
+            "document": datasets.Value(dtype="string"),
+            "text": datasets.Value(dtype="string"),
+
+            # Categories
+            "cost_type": datasets.ClassLabel(names=cost_type,),
+            "affected_entity": datasets.ClassLabel(names=affected_entity,),
+            "io_categories": datasets.Sequence(
+                datasets.ClassLabel(names=io_categories,)),
+            "aa_categories": datasets.Sequence(
+                datasets.ClassLabel(names=aa_categories,)),
+            "aa_categories_unique": datasets.Sequence(
+                datasets.ClassLabel(names=aa_categories_unique,)),
+        }
+    }
+}
+
+
+x = {
+    "config_names": {
+        "raw": {
+            "description": "",
+            "features": {
+                "source_id": datasets.Value(dtype="int64"),
+                "doc_source_id": datasets.Value(dtype="int64"),
+                "document": datasets.Value(dtype="string"),
+                "text": datasets.Value(dtype="string"),
+            }
+        }
+    }
+}
+
+BASIC_FEATURES_SPEC = {
+    "source_id": datasets.Value(dtype="int64"),
+    "doc_source_id": datasets.Value(dtype="int64"),
+    "document": datasets.Value(dtype="string"),
+    "text": datasets.Value(dtype="string"),
+}
+
+RAW_FEATURES_SPEC = {
+    "source_id": datasets.Value(dtype="int64"),
+    "doc_source_id": datasets.Value(dtype="int64"),
+    "document": datasets.Value(dtype="string"),
+    "text": datasets.Value(dtype="string"),
+}
+
+SENTENCES_UNLABELED_FEATURES_SPEC = {
+    "source_id": datasets.Value(dtype="int64"),
+    "doc_source_id": datasets.Value(dtype="int64"),
+    "document": datasets.Value(dtype="string"),
+    "text": datasets.Value(dtype="string"),
+
+    #
+
+    # Categories
+    "cost_type": datasets.ClassLabel(names=cost_type,),
+    "affected_entity": datasets.ClassLabel(names=affected_entity,),
+    "io_categories": datasets.Sequence(datasets.ClassLabel(names=io_categories,)),
+    "aa_categories": datasets.Sequence(datasets.ClassLabel(names=aa_categories,)),
+    "aa_categories_unique": datasets.Sequence(datasets.ClassLabel(names=aa_categories_unique,)),
+}
 
 
 class PY_legislation(datasets.GeneratorBasedBuilder):
@@ -29,15 +165,15 @@ class PY_legislation(datasets.GeneratorBasedBuilder):
     BUILDER_CONFIGS = [
         datasets.BuilderConfig(
             name="raw", version=VERSION,
-            description=PY_LEGISLATION_METADATA["raw-description"],
+            description=_metadata["raw"]["description"],
         ),
         datasets.BuilderConfig(
             name="sentences_unlabeled", version=VERSION,
-            description=PY_LEGISLATION_METADATA["sentences-unlabeled-description"],
+            description=_metadata["sentences-unlabeled"]["description"],
         ),
         datasets.BuilderConfig(
             name="sentences_labeled", version=VERSION,
-            description=PY_LEGISLATION_METADATA["sentences-labeled-description"],
+            description=_metadata["sentences-labeled"]["description"],
         ),
     ]
 
@@ -51,15 +187,18 @@ class PY_legislation(datasets.GeneratorBasedBuilder):
         description = ""
 
         if self.config.name == "raw":
-            description = PY_LEGISLATION_METADATA["raw-description"]
-            features = datasets.Features(RAW_FEATURES_SPEC)
+            description = _metadata["raw"]["description"]
+            features = datasets.Features(_metadata["raw"]["features"])
 
         if self.config.name == "sentences_unlabeled":
-            description = PY_LEGISLATION_METADATA["sentences-unlabeled-description"]
-            features = datasets.Features(SENTENCES_UNLABELED_FEATURES_SPEC)
+            description = _metadata["sentences-unlabeled"]["description"]
+            features = datasets.Features(
+                _metadata["sentences-unlabeled"]["features"])
 
         if self.config.name == "sentences_labeled":
-            description = PY_LEGISLATION_METADATA["sentences-labeled-description"]
+            description = _metadata["sentences-labeled"]["description"]
+            features = datasets.Features(
+                _metadata["sentences-labeled"]["features"])
 
         else:
             features = datasets.Features(
@@ -69,13 +208,14 @@ class PY_legislation(datasets.GeneratorBasedBuilder):
         return datasets.DatasetInfo(
             description=description,
             features=features,
-            homepage=PY_LEGISLATION_METADATA["homepage"],
-            license=PY_LEGISLATION_METADATA["license"],
-            citation=PY_LEGISLATION_METADATA["citation"],
+            homepage=_metadata["homepage"],
+            license=_metadata["license"],
+            citation=_metadata["citation"],
         )
 
     def _split_generators(self, dl_manager):
-        urls = PY_LEGISLATION_METADATA["urls"][self.config.name]
+        # TODO: labeled subset has two splits
+        urls = _metadata["urls"][self.config.name]
 
         urls = dl_manager.download_and_extract(urls)
 
@@ -98,4 +238,3 @@ class PY_legislation(datasets.GeneratorBasedBuilder):
                 col_name: pq_table[col_name][i].as_py()
                 for col_name in pq_table.column_names
             }
-

py_legislation_metadata.py

@@ -34,7 +34,6 @@ PY_LEGISLATION_METADATA = {
         "sentences_labeled": "./data/2_sentences_labeled/labeled.parquet",
     },
 
-    
     "raw-description" : textwrap.dedent("""
     Data extracted from the sources files (URls, PDFs and Word files) without any transformation or sentence splitter. It can be helpful because you can access the raw data extracted from the seeds (PDFs and Word files) and apply other preprocessing tasks from this point to prepare the data without returning to extract texts from source files.
     """),