ChanFormer

ChanFormer is a channel-wise Transformer architecture for multi-asset financial time-series forecasting. It is meticulously designed to study cross-asset interaction and representation learning without temporal patching, serving as a controlled architectural baseline against patch-based models.

---

Key Features

• Multi-asset modeling (crypto, equities, or blended)
• One-step-ahead OHLCV forecasting
• Channel-wise Transformer encoder
• Reversible Instance Normalization (RevIN)
• Cross-channel mixer for asset interaction
• Three experimental scenarios (crypto / stock / blended)

---

Project Structure

ChanFormer
├── README.md
├── LICENSE.txt
├── dataset/
│   ├── crypto/
│   │   └── SUIUSDT_1d_full.csv
│   └── stock/
│       └── AAPL_1d_full.csv
└── src/
    ├── ablation.py
    ├── backtesting.py
    ├── ChanFormer.py
    ├── emissions.csv
    ├── overhead.py
    ├── PatchTST.py
    └── Transformer_benchmark.py

All training, evaluation, and forecasting logic is contained in ChanFormer.py.

---

Core Idea

Instead of splitting time into patches, ChanFormer:

1. Takes a 90-day historical window
2. Flattens time per feature channel
3. Projects into embedding space
4. Applies Transformer encoding
5. Mixes channels via MLP
6. Predicts next-day OHLCV

Each feature channel becomes one token, removing temporal cross-attention and isolating the impact of channel interaction.

---

Supported Assets

#	Cryptocurrency	Stock
1	ATOMUSDT	AAPL
2	BCHUSDT	AMZN
3	DOTUSDT	AVGO
4	HBARUSDT	BRK-B
5	LTCUSDT	GOOGL
6	MATICUSDT	META
7	NEARUSDT	MSFT
8	SHIBUSDT	NVDA
9	SUIUSDT	TSLA
10	XLMUSDT	TSM

---

Model Configuration

• Sequence length: 90 days
• Embedding dimension: 384
• Attention heads: 6
• Encoder layers: 4
• Dropout: 0.10
• Optimizer: AdamW
• Learning rate: 8e-5
• Weight decay: 5e-6
• Batch size: 512
• Max epochs: 6000
• Early stopping patience: 700

---

Data Pipeline

• Timestamp alignment across assets
• Forward/backward fill missing values
• Last 365 days reserved for testing
• 20% of pre-test data for validation
• MinMax scaling per feature (fit on training only)
• Gaussian noise injection during training

---

Evaluation Metrics

Computed on scaled space:

• Global MSE
• Global RMSE
• Global MAE
• Per-asset MSE / RMSE / MAE

Predictions are inverse-transformed before saving. Results are as follows:

---

Output Files

For each scenario:

• {scenario}_forecasts_no_patch_full_ohlcv2.csv
• {scenario}_no_patch_global_result2.csv
• {scenario}_no_patch_per_asset_results.csv2
• {scenario}_model_no_patch_best.pth

---

Installation

Recommended: Python 3.10+

pip install torch pandas numpy scikit-learn

---

Usage

python ChanFormer.py

The script automatically:

• Trains all three scenarios
• Saves best model per scenario
• Exports forecasts and metrics
• Prints final performance