Traders-Lab/Preliminary-V2

Preliminary Financial Time Series Dataset (Version 2)

⚠️ Warning: This is a second preliminary dataset for development, testing, and feedback purposes. A stable, production-ready dataset will be released later. This time, all data in here will be available in future "stable" datasets (I guesss.. )

Overview

This dataset contains parquet files with time series data for various financial instruments, collected using an improved and more stable version of our data pipeline. It builds upon the first preliminary dataset, with a refined structure and more reliable data fetching processes. The data is sourced from publicly available sources like Yahoo Finance via the yfinance library.

Each financial instrument includes:

• Daily candles: Covering multiple years of historical data.
• Hourly candles: Covering at least 2 years.
• Minute candles: Covering a shorter, recent period with high granularity.

This multi-resolution format supports models that analyze both long-term trends and short-term patterns.

Purpose

This second preliminary dataset is designed to:

• Provide early access to reliable financial time series data.
• Enable testing and iteration of machine learning models for trading.
• Gather feedback to finalize a stable dataset format.
• Test the consistency of the data update pipeline over a few days.

Data Structure

The dataset is organized into a clear directory structure:

• /data/{category}/{symbol}/{symbol}.days.valid.parquet: Daily OHLC data.
• /data/{category}/{symbol}/{symbol}.hours.valid.parquet: Hourly OHLC data.
• /data/{category}/{symbol}/{symbol}.minutes.valid.parquet: Minute OHLC data.

Only files marked as .valid.parquet are included in this dataset to ensure data quality and consistency. Temporary files (e.g., fetch, test, ufetch, utest, failXX, ufailXX) are excluded via .gitignore.

Expected Changes

While the pipeline is more stable, this dataset remains preliminary. Potential changes include:

• Adjustments to file naming conventions.
• Reorganization into sharded folders (e.g., by year or month).
• Refinements to dataset split logic.

A stable, production-ready dataset will be released separately to ensure long-term consistency.

Goals

The Traders-Lab datasets aim to grow in two dimensions:

• More stocks: Additional symbols will be added over time, with rapid expansion expected soon.
• More data: Short-term datasets (hourly and minute candles) will grow as more data is accumulated. While continuity of current minute data is not guaranteed yet, future updates will ensure a continuous time history.

Non-Goals

The dataset is designed to be sufficiently up-to-date for training purposes, with data typically no more than a few days old. Real-time updates are not a goal.

License & Usage

This dataset is not licensed under a standard open data license. See the LICENSE file for detailed usage permissions. It is intended solely for research and educational purposes. Redistribution may be restricted; please respect the terms of the original data providers, such as Yahoo Finance.

Accessing the Dataset

The dataset is hosted on Hugging Face under the Traders-Lab organization. To clone the dataset:

# Ensure git-lfs is installed (https://git-lfs.com)
git lfs install
git clone https://huggingface.co/datasets/Traders-Lab/preliminary-v2

Metadata

Dataset metadata is provided in the dataset_card.yml file, following Hugging Face's dataset card standards.

Feedback

We welcome feedback to improve the dataset! Please share your thoughts via the Hugging Face Discussions or contact the Traders-Lab team.