Dataset Viewer
Auto-converted to Parquet Duplicate
seed
int64
accounts_num
int64
num_banks
int64
laundering_ratio
float64
wallet_id
dict
wallet_type
dict
wallet_level
dict
wallet_open_date
dict
wallet_open_cert_expire_offset_days
dict
bank_code_prefixes
list
normal_init_balance_mixture
list
abnormal_init_balance_mixture
list
wallet_level_balance_caps
dict
region_code
dict
distribution_based_generation
dict
output_csv
string
1,234
1,000
10
0.05
{ "length": 14, "numeric_only": true, "no_leading_zero": true }
{ "options": [ 1, 2, 3 ], "probs": [ 0.4, 0.4, 0.2 ] }
{ "options": [ 1, 2, 3, 4 ], "probs": [ 0.2, 0.05, 0.2, 0.55 ] }
{ "start": "2024-01-01T00:00:00", "end": "2025-02-18T00:00:00" }
{ "min": 365, "max": 3650 }
[ "6220", "6212", "6213", "6214", "6215", "6216", "6217", "6218", "6219", "6221" ]
[ { "type": "point", "value": 0.01, "prob": 0.01, "min": null, "max": null, "mu": null, "sigma": null }, { "type": "point", "value": 0.5, "prob": 0.01, "min": null, "max": null, "mu": null, "sigma": null }, { "type": "uniform", "value": null, ...
[ { "type": "point", "value": 0.01, "prob": 0.01, "min": null, "max": null, "mu": null, "sigma": null }, { "type": "point", "value": 0.5, "prob": 0.01, "min": null, "max": null, "mu": null, "sigma": null }, { "type": "uniform", "value": null, ...
{ "1": null, "2": 5000000, "3": 200000, "4": 100000 }
{ "codes": [ "110000", "310000", "440100", "440300", "120000", "320100", "130100", "140100", "150100", "210100", "220100", "230100", "320200", "330100", "340100", "350100", "360100", "370100", "410100", "420100", "430100", "450100...
{ "normal_wallet_distributions": { "region_code": { "normal_codes": [ "110000", "310000", "440100", "440300", "120000", "320100", "130100", "140100", "150100", "210100", "220100", "230100", "320200", ...
accounts.csv

KnowFeat: Data, Features, and Provenance Cards

This repository contains the datasets, engineered features, domain knowledge inputs, and provenance cards for the KnowFeat paper on knowledge-guided automated feature engineering with large language models.

Repository Structure

data/
  public/          # 7 public benchmark datasets (CSV)
  simecny/         # Compressed transaction data (CSV.GZ)
inputs/
  public_datasets/ # Domain knowledge JSONs for each dataset
  simecny/         # Domain knowledge for transaction data
outputs/
  features/        # Engineered feature CSV files produced by KnowFeat
  provenance/      # Provenance cards (JSON + Markdown) documenting feature lineage

Data Description

  • data/public/: Seven public tabular datasets used for evaluation (adult, bank_marketing, blood_transfusion, breast_w, credit_g, diabetes, heart).
  • data/simecny/: Proprietary transaction dataset (compressed).
  • inputs/: Domain knowledge files that guide the LLM-based feature engineering process.
  • outputs/features/: The generated features from KnowFeat and baseline methods.
  • outputs/provenance/: Provenance cards that record how each feature was created, including the LLM reasoning and statistical verification results.

Usage

These artifacts can be used to:

  1. Reproduce the experimental results from the KnowFeat paper
  2. Inspect the provenance of engineered features
  3. Benchmark new feature engineering methods against KnowFeat outputs

License

MIT

Downloads last month
350