size_categories:
- 1M<n<10M
tags:
- finance
- crypto
- BTC
Dataset Description
These datasets contain financial data related to Bitcoin. They are used to train a reinforcement learning (RL) agent whose objective is to learn how to trade the Bitcoin market.
The agent relies on an architecture inspired by the System 1 / System 2 cognitive model:
System 1: responsible for fast decision-making during trading.
System 2: a pre-trained neural network designed to analyze the market regime and provide macro-level context to improve the decisions made by System 1.
More specifically, System 2 is trained to identify the market regime for the next 24 hours and provide this information to System 1 in order to guide its trading actions.
The data is organized into two main groups of datasets: Macro-State and Micro-State.
Macro-State
The Macro-State group contains data describing the macroeconomic evolution of the Bitcoin market as well as the state of the Bitcoin network.
These datasets are used to train System 2, whose role is to detect market regimes.
The variables in these datasets are obtained from transformations applied to raw on-chain and macro-financial data, such as:
MVRV
NVT
Hash Rate
and other similar metrics.
The goal of these transformations is to extract informative signals about the market structure and dynamics, helping the agent better understand its environment.
Each row corresponds to one day of market history.
The datasets belonging to this group are:
metric_pretrain.csvmetric_train.csvmetric_test.csv
Micro-State
The Micro-State group contains data describing the short-term microeconomic dynamics of the Bitcoin market.
These datasets are used to train System 1, which performs the trading decisions.
Each row corresponds to one hour of market history.
The variables in this group are derived from calculations based on OHLCV data (Open, High, Low, Close, Volume), allowing the extraction of indicators relevant for short-term trading decisions.
The datasets belonging to this group are:
price_train.csvprice_test.csv
Dataset Time Range
The overall market history covered by the datasets spans from 2015/01/01 to 2026/03/03, but the exact time ranges vary depending on the dataset:
metric_pretrain.csv: 2015/01/01 – 2018/07/19
metric_train.csv: 2018/07/20 – 2023/05/31
metric_test.csv: 2023/06/01 – 2026/03/03
price_train.csv: 2018/07/20 – 2023/05/31
price_test.csv: 2023/06/01 – 2026/03/03