Datasets:
Download src/explicit_learning/dsl/tabledsl.py from sungguk/visual-answerability: direct link, hf CLI and curl.
- Browser
- Download file 2.42 kB
-
https://huggingface.co/datasets/sungguk/visual-answerability/resolve/main/src/explicit_learning/dsl/tabledsl.py
- Command line
-
hf download hf://datasets/sungguk/visual-answerability/src/explicit_learning/dsl/tabledsl.py
-
curl -L -o tabledsl.py https://huggingface.co/datasets/sungguk/visual-answerability/resolve/main/src/explicit_learning/dsl/tabledsl.py
2.42 kB
| """``table_dsl_v1`` operator signature table — grammar only (``docs/02`` §7, | |
| ``docs/07`` §4 P2). | |
| This third DSL covers lookups/aggregations over a table's rows and columns. P2 | |
| ships the **signature table + canonicalization profile** so the type checker and | |
| canonicalizer cover the full ``dsl`` enum and so ChartQA's underlying table | |
| (hashed in ``provenance.table_sha256``) has a target grammar. No per-source | |
| compiler is implemented here: tables are not compiled deterministically in P2. | |
| ChartQA is a C2/P4 model-compiled source — its program is proposed/verified by | |
| the dual model families, not built in code. The compiler is therefore P4. | |
| """ | |
| from __future__ import annotations | |
| from .typecheck import OpSignature | |
| # DSL marker (the ``dsl`` enum value); kept for parity with the other DSL modules. | |
| DSL = "table_dsl_v1" | |
| # --- operator signature table --------------------------------------------- | |
| # | |
| # ``COL(label)`` selects a column (scalar); ``ROW(...)`` selects a row. ``LOOKUP`` | |
| # reads one cell. ``SUM``/``AVG`` are variadic-commutative aggregations over a | |
| # column (set) of values; ``COUNT``/``EXIST`` are set selectors over a column. | |
| SIGNATURES: dict[str, OpSignature] = { | |
| "COL": OpSignature(arg_types=("string",), return_type="column", selector_kind="scalar"), | |
| "ROW": OpSignature(arg_types=("string",), return_type="row", selector_kind="scalar"), | |
| "LOOKUP": OpSignature(arg_types=("row", "column"), return_type="value", selector_kind="none"), | |
| "SUM": OpSignature( | |
| arg_types=("value",), | |
| return_type="value", | |
| commutative=True, | |
| variadic=True, | |
| selector_kind="none", | |
| ), | |
| "AVG": OpSignature( | |
| arg_types=("value",), | |
| return_type="value", | |
| commutative=True, | |
| variadic=True, | |
| selector_kind="none", | |
| ), | |
| "COUNT": OpSignature(arg_types=("column",), return_type="integer", selector_kind="set"), | |
| "EXIST": OpSignature(arg_types=("column",), return_type="boolean", selector_kind="set"), | |
| } | |
| # Canonicalization profile (§7.2 steps 3–5). ``SUM``/``AVG`` are commutative; | |
| # table aliases/casts are not introduced in P2 (the compiler is P4). | |
| COMMUTATIVE_OPS: frozenset[str] = frozenset({"SUM", "AVG"}) | |
| ALIASES: dict[str, str] = {"GET_CELL": "LOOKUP", "TOTAL": "SUM"} | |
| IDENTITY_CASTS: frozenset[str] = frozenset() | |
| __all__ = [ | |
| "ALIASES", | |
| "COMMUTATIVE_OPS", | |
| "DSL", | |
| "IDENTITY_CASTS", | |
| "SIGNATURES", | |
| ] | |