Maarij-Aqeel/market-pulse-models

Market Pulse Predictor — Sequence Models

Trained checkpoints for the RT-Market-Movement-Prediction project.

Contents

• 24 model checkpoints (checkpoints/*.pt)
  • 4 architectures (RNN, LSTM, GRU, BiLSTM-Attention) × 6 tickers (AAPL, MSFT, GOOGL, AMZN, TSLA, META)
• Feature artefacts (features/) — fitted StandardScaler, feature-name list, dataset metadata
• Model comparison (results/model_comparison.csv) — full test metrics

Inference

from huggingface_hub import hf_hub_download
import torch
import pickle

ckpt = hf_hub_download(repo_id="Maarij-Aqeel/market-pulse-models",
                       filename="checkpoints/bilstm_attention_META.pt")
scaler_path = hf_hub_download(repo_id="Maarij-Aqeel/market-pulse-models",
                              filename="features/scaler.pkl")

state_dict = torch.load(ckpt, map_location="cpu")
with open(scaler_path, "rb") as f:
    scaler = pickle.load(f)

See the GitHub repository for full inference code, training scripts, and the FastAPI/Streamlit serving stack.