TradingView Technical Analysis Vision Dataset (TVTA-Vision)

A research-grade multimodal dataset for training and evaluating Vision-Language Models on TradingView-style financial chart understanding and structured technical analysis.

Dataset Overview

The TVTA-Vision dataset provides:

• Chart images: TradingView-style candlestick charts with realistic market conditions
• Structured annotations: Technical indicators, market structure, patterns, support/resistance, trend classification, and more
• Forward performance: Historical outcome data (calculated only from future candles, preventing hindsight bias)
• QA pairs: Multiple question-answer samples per chart for instruction tuning
• Natural language analysis: Human-readable technical analysis summaries

Dataset Statistics

Split	Samples	Assets	Timeframes
Train	50	34	10
Validation	15	12	7
Test	15	15	8
Total	80	61	10

Note: This is a demonstration subset. The pipeline supports scalable generation of 50,000+ samples.

Asset Coverage

Cryptocurrencies (15 assets)

BTCUSDT, ETHUSDT, BNBUSDT, SOLUSDT, XRPUSDT, ADAUSDT, DOGEUSDT, AVAXUSDT, LINKUSDT, DOTUSDT, MATICUSDT, LTCUSDT, ATOMUSDT, UNIUSDT, ETCUSDT

US Equities (15 assets)

AAPL, MSFT, NVDA, AMZN, META, GOOGL, TSLA, AMD, NFLX, JPM, V, DIS, BA, INTC, ORCL

Indices (4 assets)

^GSPC (SPX), ^NDX (NDX), ^DJI (DJI), ^RUT (RUT)

Forex (7 assets)

EURUSD=X, GBPUSD=X, USDJPY=X, AUDUSD=X, USDCAD=X, USDCHF=X, NZDUSD=X

Commodities (4 assets)

GC=F (Gold), SI=F (Silver), CL=F (WTI Oil), BRENT (Brent Oil)

Timeframe Coverage

Timeframe	Weight	Description
1m	5%	1 minute
5m	10%	5 minutes
15m	15%	15 minutes
30m	5%	30 minutes
1h	15%	1 hour
4h	15%	4 hours
12h	5%	12 hours
1d	15%	1 day
1wk	10%	1 week
1mo	5%	1 month

Directory Structure

tradingview-technical-analysis-vision/
├── README.md                          # This file
├── dataset_card.md                    # Hugging Face Dataset Card
├── LICENSE                            # License information
├── data/
│   ├── train.jsonl                    # Training split (JSONL)
│   ├── validation.jsonl               # Validation split (JSONL)
│   └── test.jsonl                     # Test split (JSONL)
├── images/
│   ├── train/                         # Training chart images
│   ├── validation/                    # Validation chart images
│   └── test/                          # Test chart images
├── annotations/
│   ├── technical_analysis.jsonl       # Full technical analysis records
│   ├── patterns.jsonl                 # Pattern detection records
│   ├── indicators.jsonl               # Indicator state records
│   └── qa.jsonl                       # Question-answer pairs
├── metadata/
│   ├── assets.json                    # Asset universe definition
│   ├── timeframes.json                # Timeframe configuration
│   └── methodology.json               # Generation methodology
└── src/
    ├── generate_dataset.py            # Main generation pipeline
    ├── data_fetcher.py                # OHLCV data fetching (Binance, yfinance)
    ├── indicators.py                  # Technical indicator calculations
    ├── chart_generator.py             # TradingView-style chart image generation
    ├── patterns.py                    # Pattern and structure detection
    └── forward_performance.py         # Forward performance calculation

Data Schema

Each JSONL record contains:

{
  "id": "tvta_000001",
  "image": "images/train/tvta_000001.png",
  "image_source_type": "tradingview_style_generated",
  "data_source": "real_binance | real_yfinance | synthetic",
  "asset": {
    "symbol": "BTCUSDT",
    "asset_class": "cryptocurrency"
  },
  "timeframe": "4h",
  "timestamp": "2024-05-12T12:00:00Z",
  "chart_window": {
    "start_idx": 400,
    "end_idx": 460,
    "total_candles": 480
  },
  "ohlcv": {
    "open": 60000.0,
    "high": 61200.0,
    "low": 59500.0,
    "close": 60800.0,
    "volume": 28450.0
  },
  "indicators": {
    "rsi": {"value": 64.2, "state": "bullish_momentum", "zone": "neutral_bullish"},
    "macd": {"value": 425.2, "signal": 381.4, "histogram": 43.8, "state": "bullish"},
    "ema_structure": {"ema20_above_ema50": true, "ema50_above_ema200": true, "alignment": "bullish"},
    "adx": {"value": 28.5, "trend_strength": "strong"},
    "volume_analysis": {"current_volume_vs_average": 1.82, "classification": "high_volume"}
  },
  "market_structure": {
    "classification": "bullish",
    "sequence": "HH_HL_HH_HL",
    "break_of_structure": true,
    "change_of_character": false
  },
  "trend": {
    "classification": "bullish",
    "score": 6,
    "trending": true,
    "adx": 28.5
  },
  "patterns": [
    {"pattern": "ascending_triangle", "strength": "medium"}
  ],
  "candlestick_patterns": [
    {"pattern": "bullish_engulfing", "index": 458, "strength": "strong"}
  ],
  "support_resistance": {
    "support_levels": [{"price": 59500.0, "strength": "strong", "touches": 3}],
    "resistance_levels": [{"price": 61500.0, "strength": "medium", "touches": 2}]
  },
  "divergence": [],
  "breakout_analysis": {
    "status": "confirmed_breakout",
    "direction": "up",
    "level": 61000.0,
    "volume_confirmation": true
  },
  "market_regime": {
    "primary": "bull_trend",
    "secondary": ["breakout_regime"],
    "volatility_annualized": 0.45
  },
  "technical_bias": {
    "score": 7,
    "bias": "bullish",
    "breakdown": {"trend": 2, "market_structure": 2, "momentum": 1, "volume": 1, "moving_averages": 1}
  },
  "technical_confidence": 0.81,
  "forward_performance": {
    "return_1c": 0.004,
    "return_5c": 0.021,
    "return_10c": 0.037,
    "return_20c": 0.012,
    "max_favorable_excursion": 0.045,
    "max_adverse_excursion": -0.008,
    "max_drawdown": -0.012,
    "future_volatility": 0.38
  },
  "signal_outcome": "confirmed_breakout",
  "difficulty": "medium",
  "split": "train",
  "qa": [
    {
      "question": "What is the dominant trend shown on this chart?",
      "answer": "Bullish",
      "type": "trend_identification"
    }
  ],
  "analysis": "BTCUSDT on the 4h timeframe shows a bullish trend..."
}

Quick Start

Install Dependencies

pip install pandas numpy matplotlib yfinance mplfinance

Generate Dataset

# Generate training split
python src/generate_dataset.py --samples 1000 --output data/train.jsonl --split train

# Generate validation split
python src/generate_dataset.py --samples 150 --output data/validation.jsonl --split validation

# Generate test split
python src/generate_dataset.py --output data/test.jsonl --split test --samples 150

Load Dataset

import json

# Load JSONL
records = []
with open("data/train.jsonl") as f:
    for line in f:
        records.append(json.loads(line))

# Access fields
for record in records[:5]:
    print(f"{record['asset']['symbol']} | {record['timeframe']}")
    print(f"Trend: {record['trend']['classification']}")
    print(f"Bias: {record['technical_bias']['bias']}")
    print(f"Confidence: {record['technical_confidence']}")
    print(f"Forward 10c return: {record['forward_performance'].get('return_10c', 'N/A')}")
    print()

Data Sources

• Cryptocurrency OHLCV: Binance Public API (real-time historical data)
• Equity/Index/Forex/Commodity OHLCV: Yahoo Finance via yfinance
• Synthetic fallback: Geometric Brownian Motion with asset-specific parameters (clearly labeled)

Anti-Hindsight Design

The dataset strictly separates:

1. Technical analysis labels: Computed ONLY from data visible at or before the chart's final candle
2. Forward performance: Computed ONLY from future candles after the chart's final candle

This prevents future information from contaminating the technical analysis labels.

Quality Control

Every sample passes automated validation:

• OHLC consistency (High >= Open/Close, Low <= Open/Close)
• Volume >= 0
• RSI range 0-100
• Valid timestamps and asset symbols
• No future leakage in technical labels
• Correct forward-return calculations

Intended Use

• Multimodal financial AI research
• Chart understanding model training
• Technical analysis automation
• Vision-language model evaluation
• Financial chart classification

Out-of-Scope Use

This dataset is for research and model training only. It should NOT be treated as guaranteed investment advice. Technical analysis patterns do not guarantee future price movements.

License

See LICENSE for details.

• OHLCV data: From Binance and Yahoo Finance (subject to their terms)
• Chart images: Generated (no copyright claims on visual style)
• Annotations: MIT License

Citation

@dataset{tvta_vision_2024,
  title={TradingView Technical Analysis Vision Dataset (TVTA-Vision)},
  author={Dataset Engineering Pipeline},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/your-username/tvta-vision}
}

Acknowledgments

• TradingView for chart style inspiration
• Binance for cryptocurrency market data
• Yahoo Finance for equity/forex/commodity data