| --- |
| license: mit |
| size_categories: |
| - 100M<n<1B |
| task_categories: |
| - time-series-forecasting |
| - tabular-classification |
| pretty_name: AMM-Events (Event-Aware DeFi Dataset) |
| tags: |
| - finance |
| - defi |
| - amm |
| - ethereum |
| - cryptocurrency |
| - transaction |
| - microstructure |
| --- |
| |
| # AMM-Events: A Multi-Protocol DeFi Event Dataset |
|
|
| ## Dataset Description |
|
|
| **AMM-Events** is a high-fidelity, block-level dataset capturing **8.9 million on-chain events** from the Ethereum mainnet, specifically designed for event-aware forecasting and market microstructure analysis in Decentralized Finance (DeFi). |
|
|
| Unlike traditional financial datasets based on Limit Order Books (LOB), this dataset focuses on **Automated Market Makers (AMMs)**, where price dynamics are triggered exclusively by discrete on-chain events (e.g., swaps, mints, burns) rather than continuous off-chain information. |
|
|
| - **Paper:** [Towards Event-Aware Forecasting in DeFi: Insights from On-chain Automated Market Maker Protocols](https://huggingface.co/papers/2604.20374) |
| - **Code:** [https://github.com/yosen-king/Deep-AMM-Events](https://github.com/yosen-king/Deep-AMM-Events) |
| - **Total Events:** 8,917,353 |
| - **Time Span:** Jan 1, 2024 – Sep 16, 2025 |
| - **Block Range:** 18,908,896 – 23,374,292 |
| - **Protocols:** Uniswap V3, Aave, Morpho, Pendle |
| - **Granularity:** Block-level timestamps & transaction-level event types |
|
|
| ### Supported Tasks |
| - **Event Forecasting:** Predicting the next event type (classification/TPP) and time-to-next-event (regression/TPP). |
| - **Market Microstructure Analysis:** Analyzing causal synchronization between liquidity events and price shocks. |
| - **Anomaly Detection:** Identifying "Black Swan" traffic surges or congestion events. |
|
|
| --- |
|
|
| ## Dataset Structure |
|
|
| The data is organized into a standardized JSON format. Each entry decouples complex smart contract logic into interpretable metrics. |
|
|
| ### Data Fields |
|
|
| - `block_number` (int): The Ethereum block height where the event occurred. |
| - `timestamp` (int): Unix timestamp of the block. |
| - `transaction_hash` (string): Unique identifier for the transaction. |
| - `protocol` (string): Origin protocol (`Uniswap V3`, `Aave`, `Morpho`, or `Pendle`). |
| - `event_type` (string): The category of the event (`Swap`, `Mint`, `Burn`, `UpdateImpliedRate`, etc.). |
| - `payload` (dict): Protocol-specific metrics (e.g., `amount0`, `amount1`, `liquidity`, `tick` for Uniswap). |
|
|
| ### Data Splits |
|
|
| The dataset covers **359 liquidity pools** selected for high activity and representativeness: |
| - **Pendle:** 296 pools (Yield Trading) |
| - **Aave:** 53 pools (Lending) |
| - **Uniswap V3:** 5 pools (Spot Trading) |
| - **Morpho:** 5 pools (Lending Optimization) |
|
|
| --- |
|
|
| ## Usage |
|
|
| ### Loading the Data |
| You can load this dataset directly using the Hugging Face `datasets` library: |
|
|
| ```python |
| from datasets import load_dataset |
| |
| dataset = load_dataset("Jackson668/AMM-Events") |
| |
| # Example: Accessing the first train example |
| print(dataset['train'][0]) |
| ``` |
