# Alpha Radar: Solana Sprint — Data

来源：https://www.kaggle.com/competitions/alpha-radar-solana-sprint/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

This dataset contains the first 30 seconds of post-mint transaction data for PumpFun tokens created in the month of September 2025.
The data was scraped using a proprietary parsing tool from a private Solana node.
The categories are fixed and must be adhered to.

**Data Sources**

The full 30-second dataset:
🔗 https://www.kaggle.com/datasets/twainayar/pumpfun-30s-september-2025

The target tokens (wallet buys):
🔗 https://drive.google.com/file/d/1EsqpZXPBU-6m0djDmccCrtUX07jV2fHA/view?usp=sharing

**The Goal**

The target tokens are contained within the larger dataset.
Using the available categories, engineer a CatBoost model that can isolate at least 75 % of the target tokens within the first 30 seconds of their lifespan.

**Update (October 2025)**

A new evaluation dataset has been uploaded to prevent data leakage and improve leaderboard accuracy.
All submissions must now contain exactly 64 208 rows, one per token ID in the evaluation set.

Columns

**index**	Row ID / sequence
**timestamp**	Event time (UTC)
**mint_token_id**	Token mint address
**holder**	Wallet address involved
**trade_mode**	buy / sell / other action type
**token_quantity**	Tokens bought / sold amount
**creator**	Token creator address
**creator_fee**	Creator fee (absolute)
**creator_fee_pump**	PumpFun creator fee
**market_cap_usd**	Circulating market cap (USD)
**token_delta**	Net token change
**sol_delta**	Net SOL spent / received
**buy_count**	Number of buys
**sell_count**	Number of sells
**total_count**	Total transactions (buys + sells)
**token_volume**	Traded token volume
**sol_volume**	Traded SOL volume
**liquidity_ratio**	Liquidity / volume ratio
**virtual_sol_reserves**	Virtual SOL in pool
**virtual_token_reserves**	Virtual tokens in pool
**consumed_gas**	Compute units used
**fee**	Transaction fee in SOL
**relative_strength_index**	RSI momentum indicator
**bollinger_relative_position**	Price vs Bollinger bands
**volume_oscillator**	Volume momentum measure
**rate_of_change**	% change over window
**money_flow_index**	Volume-weighted RSI
**total_holders**	All-time unique holders
**current_holders**	Current wallets holding
**top10_percent_total**	Top 10 % holdings share
**creator_balance**	Creator’s current token balance
**creator_sold**	Tokens sold by creator
**holder_ratio**	Holders / supply metric
**buy_sell_ratio**	Buys / sells ratio

**Prediction Target**

Your objective is to predict whether a token appears in the Target Token Dataset.
This is a binary classification task.

Target label:

is_target = 1 → token was bought by the target wallet  
is_target = 0 → token was not bought

Submission Format

Your submission must be a CSV with the following structure:

mint_token_id,is_target
<MINT_TOKEN_ID>,<IS_TARGET>


**Example:**

mint_token_id,is_target
14XBCQLJ6A8jifCBFsp1NfW3kG9FxbY7xbiQSxfXpump,1
4shdnB4wA5TA5wphgD1gQFmbaAACtbUyiepLDUoopump,0


**Header row required
Exactly 64 208 rows (one for each token in the evaluation set)
is_target values must be binary (0 or 1)**