Datasets:
license: gpl-3.0
task_categories:
- time-series-forecasting
pretty_name: BigStock
dataset_info:
features:
- name: timestamp
dtype: string
- name: ticker
dtype: string
- name: open
dtype: float64
- name: high
dtype: float64
- name: low
dtype: float64
- name: close
dtype: float64
- name: volume
dtype: int64
- name: source
dtype: string
- name: retreived
dtype: string
splits:
- name: train
num_bytes: 15162994840
num_examples: 123911229
download_size: 4535255991
dataset_size: 15162994840
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
tags:
- timeseries
- economics
BigStock
This dataset contains historical data stock data spanning a period of 720+ days, at one-hour intervals. All values are in USD. It includes full candlestick data (open, close, low, high) as well as volume data. The data (not scraping/processing code) is licensed under GPL v3.
Other Datasets
A sample dataset (<20MB) for a subset of popular stocks is available as well. I may also release a multivariate forecasting model trained on the data, likely based on Tiny Time Mixers or a novel architecture I have been evaluating.
Uses
The data is usable for the following cluster of tasks:
- Time series forecasting
- Financial modeling
- Reinforcement Learning
Limitations
While its breadth is very large, the data is comparatively very short-term (~2yrs). Controlling for biases in economic data is inherently a difficult problem, because we only have one market to sample from. Therefore, models trained on this data may be biased towards recent market trends.
This dataset may have missing, incorrect, or unreliable data. You are encouraged to validate it before use.
Dataset Creation
The source data in this sample was provided purely from the Yahoo Finance website. Data was collected using the yfinance library.
The collection and processing was done on hardware owned and operated entirely by me.