At a high level it provides the following major components: A full orderbook granularity, stochastic process and functions for synthetic data generation, Distributed convex methods for model training/inference.
From the standard representation (level_price, level_volume) Levels, to a by the level order-queue granularity (level_orders [ (order_id, order_price, order_amount), (order_id, order_price, order_amount), ... ] ) to provide a true order-driven market representation structure for enriched models.
Stochastic process generators for rich/complex simulations, implementations include: Uniform, Brownian, Hawkes.
e## Distributed Convex Methods
Distributed Convex Methods for Linear Models Training and inference. Implementations include: Undirected Acyclic Compute Graph with Gradient Consensus.
Clone the repository
git clone
cd atelier
cargo run \
-- --template "atelier-sync/templates/single_orderbook.toml" \
--output-dir "./examples"If you are using a mac with Apple sillicon, you just need to build with --platform linux/amd64 in order to cross compile, within the OSx system, the linux vm in the container, otherwise just do not include it.
docker build \
--platform linux/amd64 \
--target runner \
--file .Dockerfile \
--tag atelier-torch \
--no-cache . the builder stage, to compile the rust binary, and the runner stage to have a
minimalistic container to expose a service provided by the binary execution.
Generating results by running the containerized atelier.
docker run \
--platform linux/amd64 \
atelier-torch \
--template "templates/single_orderbook.toml" \
--output-dir "."These are the other published crates members of the workspace:
- atelier-data: Core data structures and I/O tools.
- atelier-dcml: Distributed Convex Machine Learning.
- atelier-generators: Probabilistic generators and events simulation.
- atelier-results: Standardized results, errors and successes.
- atelier-synth: Synthetic Data Generation for the atelier-rs engine.
Github hosted:
This project is licensed under the Apache V2 license. Any contribution intentionally submitted for inclussion in Atelier by you, shall be licensed as Apache V2, without any additional terms or conditions.