Add dataset card

README.md

@@ -1,37 +1,112 @@
 ---
-dataset_info:
-  features:
-  - name: index
-    dtype: int64
-  - name: statement_id
-    dtype: int64
-  - name: statement
-    dtype: string
-  - name: lineno
-    dtype: int64
-  - name: end_lineno
-    dtype: int64
-  - name: comprehension_types
-    sequence: string
-  - name: metadata
-    struct:
-    - name: hexsha
-      dtype: string
-    - name: max_stars_repo_name
-      dtype: string
-    - name: max_stars_repo_path
-      dtype: string
-    - name: lang
-      dtype: string
-  splits:
-  - name: train
-    num_bytes: 266226
-    num_examples: 1000
-  download_size: 78285
-  dataset_size: 266226
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
+language:
+- code
+- en
+tags:
+- code
+- python
+- comprehension
+- the-stack
+pretty_name: Python Comprehension Statements 1K
+size_categories:
+- 1K<n<10K
+task_categories:
+- text-generation
+- fill-mask
+license: other
 ---
+
+# Python Comprehension Statements 1K
+
+This dataset contains 1,000 Python statements extracted from `bigcode/the-stack`
+that include at least one Python comprehension expression:
+
+- list comprehensions
+- set comprehensions
+- dict comprehensions
+- generator expressions
+
+Each row is a nearest enclosing Python statement around one or more
+comprehension expressions, along with line number information and source
+metadata copied from the original The Stack sample.
+
+## Dataset Details
+
+- **Dataset name:** `py_comprehension_statements_1k`
+- **Rows:** 1,000
+- **Split:** `train`
+- **Source dataset:** `bigcode/the-stack`, Python subset
+- **Extraction script:** `scripts/collect_stack_comprehension_statements.py`
+- **Original local JSONL:** `outputs/the_stack_python_comprehension_statements.jsonl`
+
+## Schema
+
+| Field | Type | Description |
+| --- | --- | --- |
+| `index` | `int64` | Zero-based index of the streamed source sample. |
+| `statement_id` | `int64` | Statement index within the source sample. |
+| `statement` | `string` | Extracted Python statement containing a comprehension. |
+| `lineno` | `int64` | Starting line number in the original source file. |
+| `end_lineno` | `int64` | Ending line number in the original source file. |
+| `comprehension_types` | `list[string]` | AST comprehension node types present in the statement. |
+| `metadata.hexsha` | `string` | Source commit hash from The Stack metadata. |
+| `metadata.max_stars_repo_name` | `string` | Repository name from The Stack metadata. |
+| `metadata.max_stars_repo_path` | `string` | File path from The Stack metadata. |
+| `metadata.lang` | `string` | Source language label. |
+
+## Comprehension Type Counts
+
+The `comprehension_types` field can contain multiple values for a single row.
+Across the 1,000 rows, the type occurrences are:
+
+| Type | Count |
+| --- | ---: |
+| `ListComp` | 754 |
+| `GeneratorExp` | 181 |
+| `DictComp` | 76 |
+| `SetComp` | 9 |
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("nexround/py_comprehension_statements_1k")
+train = dataset["train"]
+
+print(train[0]["statement"])
+print(train[0]["comprehension_types"])
+```
+
+If you have the dataset saved locally with `datasets.save_to_disk`, use:
+
+```python
+from datasets import load_from_disk
+
+dataset = load_from_disk("outputs/the_stack_python_comprehension_statements_hf_dataset")
+```
+
+## Construction
+
+The extraction process streamed Python files from The Stack, parsed each source
+file with Python's `ast` module, found `ListComp`, `SetComp`, `DictComp`, and
+`GeneratorExp` nodes, and saved the nearest enclosing statement for each unique
+statement span.
+
+Rows that could not be parsed as Python source were skipped.
+
+## Intended Uses
+
+This dataset is intended for lightweight analysis and experimentation around
+Python comprehension syntax, such as:
+
+- inspecting comprehension usage patterns
+- evaluating code models on comprehension-heavy snippets
+- constructing small probes or prompts involving Python comprehensions
+
+## Licensing
+
+This is a derived sample from The Stack. The original source code snippets come
+from many upstream repositories with varying licenses. Users should consult and
+respect the licensing metadata and terms of the upstream The Stack dataset and
+the original repositories when using or redistributing examples.