tecnologiactc/automl_llm_agent_m1

AutoML-LLM Agent Module 1 Benchmark

This dataset contains the Module 1 benchmark for evaluating an AutoML assistant that interprets user requests, selects tabular modeling settings, and produces an auditable AutoGluon Tabular plan.

The repository is scoped to Module 1 only.

Tables

• cases: one row per Module 1 evaluation case.
• queries: one user request per case.
• columns_annotations: column-level reference annotations for target, feature, exclusion, leakage, preprocessing, and review decisions.
• module1_eval_long: denormalized table for quick inspection in the Hugging Face Dataset Viewer.
• dataset_inventory: inventory of the local CSV assets used by the Module 1 cases.
• ablation_variants: LangGraph node-ablation variants used by Module 1 experiments.

File Assets

• input_data/: raw and processed CSV files referenced by the Module 1 cases.
• metadata/M1_reference_output.csv: original reference file used as the starting point.
• output_schema.json: public Module 1 output contract, including audit artifacts.

Current Coverage

The benchmark currently includes seven reference cases:

• calls_for_service_original
• electric_vehicle_original
• cholesterol_original
• diabetes_original
• properties_original
• banking_original
• avocado_original

Some cases intentionally document pending or manual-review gaps. In particular, avocado_original is present as a planned case while its CSV files are not yet available locally, and calls_for_service_original is marked for manual review because the requested target differs from the available raw columns.

Usage

from datasets import load_dataset

cases = load_dataset("tecnologiactc/automl_llm_agent_m1", "cases", split="train")
annotations = load_dataset("tecnologiactc/automl_llm_agent_m1", "columns_annotations", split="train")
viewer = load_dataset("tecnologiactc/automl_llm_agent_m1", "module1_eval_long", split="train")

Raw CSV assets referenced by cases.input_data_path can be downloaded from the same dataset repository.