Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -17,6 +17,16 @@ tags:
|
|
| 17 |
|
| 18 |
A multimodal financial analysis model that combines textual market sentiment with visual candlestick patterns for enhanced trading signal prediction and price forecasting.
|
| 19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
## Architecture Overview
|
| 21 |
|
| 22 |
### Core Components
|
|
@@ -39,15 +49,16 @@ A multimodal financial analysis model that combines textual market sentiment wit
|
|
| 39 |
- **Input Text**: Tokenized to max 64 tokens
|
| 40 |
- **Input Images**: Resized to 224x224 RGB
|
| 41 |
- **Hidden Dimension**: 768 (consistent across encoders)
|
| 42 |
-
- **Output Classes**:
|
| 43 |
- **Dropout**: 0.3 in both heads
|
| 44 |
|
| 45 |
## Training Details
|
| 46 |
-
- **Epochs**:
|
| 47 |
- **Learning Rate**: 2e-05
|
| 48 |
- **Loss Function**: CrossEntropy (classification) + MSE (regression)
|
| 49 |
- **Loss Weight (alpha)**: 0.5 for regression term
|
| 50 |
- **Optimizer**: AdamW with linear scheduling
|
|
|
|
| 51 |
|
| 52 |
## Usage
|
| 53 |
```python
|
|
|
|
| 17 |
|
| 18 |
A multimodal financial analysis model that combines textual market sentiment with visual candlestick patterns for enhanced trading signal prediction and price forecasting.
|
| 19 |
|
| 20 |
+
## Links
|
| 21 |
+
- 🔗 **GitHub Repository**: https://github.com/tuankg1028/CandleFusion
|
| 22 |
+
- 🚀 **Demo on Hugging Face Spaces**: https://huggingface.co/spaces/tuankg1028/candlefusion
|
| 23 |
+
|
| 24 |
+
## Training Results
|
| 25 |
+
- **Best Epoch**: 18
|
| 26 |
+
- **Best Validation Loss**: 463938.7056
|
| 27 |
+
- **Training Epochs**: 23
|
| 28 |
+
- **Early Stopping**: Yes
|
| 29 |
+
|
| 30 |
## Architecture Overview
|
| 31 |
|
| 32 |
### Core Components
|
|
|
|
| 49 |
- **Input Text**: Tokenized to max 64 tokens
|
| 50 |
- **Input Images**: Resized to 224x224 RGB
|
| 51 |
- **Hidden Dimension**: 768 (consistent across encoders)
|
| 52 |
+
- **Output Classes**: 3 (buy/sell/hold)
|
| 53 |
- **Dropout**: 0.3 in both heads
|
| 54 |
|
| 55 |
## Training Details
|
| 56 |
+
- **Epochs**: 100
|
| 57 |
- **Learning Rate**: 2e-05
|
| 58 |
- **Loss Function**: CrossEntropy (classification) + MSE (regression)
|
| 59 |
- **Loss Weight (alpha)**: 0.5 for regression term
|
| 60 |
- **Optimizer**: AdamW with linear scheduling
|
| 61 |
+
- **Early Stopping Patience**: 5
|
| 62 |
|
| 63 |
## Usage
|
| 64 |
```python
|