---
license: mit
tags:
- finance
- stock-market
- indian-market
- nse
- nifty
- time-series
- transformer
- investment
datasets:
- xxparthparekhxx/indian-stock-market-minute-data
pipeline_tag: tabular-classification
---

# 🇮🇳 Indian Stock Market Analyzer

**Multi-task Transformer for Indian stock market analysis and real-time investment decisions.**

## Architecture
- Transformer encoder (4 layers, 8 heads, d=128) with causal attention
- Wavelet-denoised OHLCV + 14 technical indicators (RSI, MACD, Bollinger, ATR, etc.)
- Multi-task: direction (BUY/HOLD/SELL) + 5-day return + confidence calibration
- Based on: TIPS, MASTER, xLSTM-TS research papers

## Performance
| Metric | Value |
|--------|-------|
| Direction Accuracy | 41.98% |
| F1 Score (macro) | 41.62% |
| Return IC | 0.3582 |
| Strategy Sharpe | 2.193 |
| Annualized Return | 114.48% |

## Stocks: Top 50 NSE stocks
IDEA, TATASTEEL, YESBANK, ETERNAL, PCJEWELLER, ADANIPOWER, PRIVISCL, EASEMYTRIP, SUZLON, IRFC, RELIANCE, NMDC, IREDA, ASHOKLEY, IDFCFIRSTB...

## Usage
```python
import torch, json
config = json.load(open('config.json'))
# Load StockTransformerAnalyzer model
model.load_state_dict(torch.load('model.pt'))
# Input: (batch, 30 days, 14 features) → Output: BUY/HOLD/SELL + return + confidence
```

## ⚠️ Disclaimer
For research/education only. NOT financial advice. Consult a financial advisor.