chestnutforty/forecast-snapshots-metaculus-6f1cdfd9b3

Forecast Snapshot Dataset

Source: metaculus Data Hash: 6f1cdfd9b3

Dataset Description

This dataset contains snapshots of prediction market data for AI agent evaluation.

Format

Each row represents a market snapshot at a specific point in time, including:

• —Current community prediction
• —Predictions 1 day, 3 days, 1 week, 2 weeks, 1 month later (or resolution)
• —Final resolution (ground truth)
• —Market metadata (question, close time, etc.)

Question Types

• —binary: Yes/No questions (scalar predictions)
• —multiple_choice: Categorical options (array of probabilities for all options)
• —numeric/discrete/date: Continuous predictions with percentiles

Note: In the Parquet file, arrays are stored as JSON strings. In the CSV file, arrays are stored as Python list representations.

Usage

import pandas as pd

# Load dataset (CSV or Parquet)
df = pd.read_csv("hf://datasets/chestnutforty/forecast-snapshots-metaculus-6f1cdfd9b3/snapshot_dataset.csv")
# or
df = pd.read_parquet("hf://datasets/chestnutforty/forecast-snapshots-metaculus-6f1cdfd9b3/snapshot_dataset.parquet")

# Filter to specific snapshot date
snapshot_date = "2025-01-01T00:00:00"
markets = df[df['snapshot_datetime'] == snapshot_date]

# Evaluate model predictions
# df['model_pred_now'] = your_model.predict(markets)

Configuration

See config.json for dataset creation parameters.

Generated with Claude Code