ml-quant-trading synthetic MLP smoke-test checkpoint
This is the small MLPRegressor checkpoint produced by the deterministic
mlquant demo pipeline. It is published to make installation and inference
paths reproducible — not as a market model or investment signal.
Quick start
Install the verified package from PyPI before loading this checkpoint:
python -m pip install --upgrade mlquantx huggingface_hub
mlquant demo
The distribution name is mlquantx, while Python imports and the CLI use mlquant. The full source, tests, research notes, and the v0.2.6 release are available in the GitHub repository and release page.
The checkpoint is a reproducible synthetic research artifact. It is not evidence of live trading performance.
Associated paper
This artifact accompanies
Machine Learning Enhanced Multi-Factor Quantitative Trading
(arXiv:2507.07107).
Model details
- Architecture: three-layer MLP with GELU and dropout
- Input dimensions:
213factor values - Hidden width:
64 - Output: one synthetic next-period return score per asset-date
- Training data:
dddyym/ml-quant-trading-synthetic - Source commit:
faababb851b22061759f748c252f1cca1eaf0202 - Checkpoint SHA-256:
57a2773f4f928ddf81660f33d3d9ef5ae2e6a9958e0c20694b3cf1234968fb84
The companion feature_names.json, config.json, source_config.yaml, and
metrics.json files define the input order, architecture, generator settings,
and synthetic smoke-test output.
Load
import json
import torch
from huggingface_hub import hf_hub_download
from mlquant.models.nets import MLPRegressor
repo_id = "dddyym/ml-quant-trading-synthetic-mlp"
config_path = hf_hub_download(repo_id, "config.json")
weights_path = hf_hub_download(repo_id, "pytorch_model.bin")
config = json.load(open(config_path))
model = MLPRegressor(
in_dim=config["in_dim"],
hidden=config["hidden"],
dropout=config["dropout"],
)
model.load_state_dict(torch.load(weights_path, map_location="cpu", weights_only=True))
model.eval()
Limitations
- The checkpoint was trained only on deterministic synthetic data.
- Synthetic smoke-test metrics are not evidence of out-of-sample alpha.
- Factor values must use the exact order in
feature_names.json. - The package is research software, not a live-trading system.
See the project's Research Card and Reality Check.