Hugging Face's logo

Datasets:
lyang84
/
UserModeling

Modalities:
Tabular
Text
Formats:
parquet
Size:
1M - 10M
Libraries:
Datasets
Dask
Polars
Dataset card Data Studio Files
xet
 Community
1
UserModeling / manifold /dataset /README_FILES.md
lyang84's picture
lyang84
add manifold benchmark dataset (belief+trade prepared splits/support)
e6b8ddf
preview code
|
Raw
History Blame Contribute Delete
3.65 kB

Dataset File Guide (Manifold Benchmark)

This document explains what each file in dataset/ is for, including keys and common usage.

Common IDs

  • userId: Manifold user ID
  • marketId: Manifold market/contract ID
  • betId: Unique bet event ID
  • answerId: Candidate ID for multi-choice markets

Belief

Belief/source/data.csv

Granularity: one row per userId * marketId

Contains final belief aggregation for each user-market pair.

Important columns:

  • dominant_option: top accumulated option for this user-market
  • dominant_net: net accumulated amount on dominant option
  • dominant_share: |dominant_net| / total_abs_net
  • candidate_options_json, candidate_net_json, candidate_bets_json: full option-level breakdown

Use this as the main target table for Belief prediction.

Belief/source/option_accumulate.csv

Granularity: one row per userId * marketId * option_key

Option-level accumulate details.

Important columns:

  • option_key: ANS:<answerId> for multi-choice, SIDE:YES/NO for binary-like
  • accumulate_net_amount: signed net accumulate amount for that option
  • bets_count: number of events contributing to that option

Use this when you need detailed candidate-level signals.

Belief/splits/train.csv, validation.csv, test.csv

Granularity: one row per userId * marketId

Query split files for Belief tasks. Built by per-user chronological splitting (market-level timestamp ordering), then merged globally.

Belief/splits/wallet_ids_experiment_2000.csv

The 2000 sampled query users used for experiment splits.

Belief/splits/split_summary.txt

Summary counts for train/validation/test.

Belief/support_set/support_users.csv

Support user list and user-level stats.

Belief/support_set/support_belief_final.csv

Support users' user*market final belief summary (dominant/candidate aggregate level).

Belief/support_set/support_belief_option_accumulate.csv

Support users' candidate-level accumulate rows.

Belief/support_set/support_user_bets_chrono.csv

Support users' deduplicated bet events sorted by:

  1. userId, 2) createdTime, 3) betId.

Useful for retrieval/profile construction.

Belief/support_set/support_user_bets_index.csv

Row-range index for support_user_bets_chrono.csv.

Columns:

  • start_row_in_events_csv_1based_excluding_header
  • end_row_in_events_csv_1based_excluding_header
  • num_bets

Use this to quickly retrieve a single user's full event history.

Trade

Trade/splits/train.csv, validation.csv, test.csv

Granularity: one row per bet event (betId)

Query trade split files. For each query user, bets are sorted by createdTime and split by 85/10/5 chronologically, then merged.

Important columns include:

  • userId, createdTime, betId, marketId
  • outcome, answerId
  • amount, shares
  • probBefore, probAfter
  • isApi

Trade/splits/split_summary.txt

Summary counts for Trade splits.

Trade/support_set/support_trades.csv

Support users' deduplicated trade events in chronological order per user.

Trade/support_set/support_trades_index.csv

Per-user row-range index into support_trades.csv.

Trade/support_set/support_summary.txt

Support user/bet count summary.

Recommended Join Paths

  • Belief query rows (Belief/splits/*.csv) ↔ source belief (Belief/source/data.csv) by (userId, marketId).
  • Trade query rows (Trade/splits/*.csv) are already event-level; join to support by userId if needed.
  • For multi-choice text reconstruction, map answerId using market metadata/API snapshots.