hajirazin/learnfinance-models-patchtst

LearnFinance PatchTST Model - v2026-05-08-d40d9199c8bb

OHLCV 5-channel PatchTST transformer model for predicting weekly stock returns.

Model Details

• Version: v2026-05-08-d40d9199c8bb
• Model Type: PatchTST (Patch Time Series Transformer)
• Training Window: 2016-01-01 to 2026-05-08
• Symbols: 434 stocks

Input Channels (11 total)

• OHLCV log returns (5): open, high, low, close, volume
• News sentiment (1)
• Fundamentals (5): gross_margin, operating_margin, net_margin, current_ratio, debt_to_equity

Metrics

• Train Loss: 0.07492728165145647
• Validation Loss: 0.08409886006371113
• Baseline Loss: 0.08807823807001114

Usage

from brain_api.storage.patchtst.huggingface import PatchTSTHalalNewHuggingFaceModelStorage
from brain_api.storage.patchtst.local import PatchTSTHalalNewModelStorage

storage = PatchTSTHalalNewHuggingFaceModelStorage(
    repo_id="hajirazin/learnfinance-models-patchtst",
    local_cache=PatchTSTHalalNewModelStorage(),
)
artifacts = storage.download_model(version="v2026-05-08-d40d9199c8bb")