How yukikihan/mfb works
Project type and purpose
• This is a data‐science/ML research pipeline implemented in Python via Jupyter notebooks.
• Its goal is to predict short‐term Bitcoin price movements using a multimodal fusion (MFB) model that leverages time‐lagged sentiment from tweets and news plus traditional market indicators.
File structure
• Top level: several .ipynb notebooks for each experiment stage, a utils.py helper, a LICENSE and README.md.
• A dataset/ directory containing raw data (crypto/price.csv, ti.csv) and processed outputs (processed/*.csv).
• The notebooks are numbered by workflow step: Data Standardization, Correlation Analysis, Feature Validation, Data Split, Baselines & MFB on tweets/news, Optimizer Comparison, Final Price Prediction.
README for additional insights
• Workflow steps are clearly enumerated (data collection → preprocessing → correlation → feature optimization → split → training & evaluation → forecasting).
• External data sources: Kaggle tweets, News RESTful API, CoinMarketCap.
• Core technology: Python, VADER sentiment, custom MFB model, PyCharm for IDE.
Elements
• Rectangles for processing modules (notebooks)
• Hexagons or clouds for external APIs
• Parallelograms for output reports/forecasts