How syt20/deep-learning-cryptocurrency-forecasting-engine works
Project Type and Purpose
• This is a data-science / exploratory forecasting tool built around a Jupyter Notebook (BitScry.ipynb).
File Structure
• Top-level:
• BitScry.ipynb – the main orchestration and user interface for experiments.
• Data/bitcoin.csv – raw input data source.
• Experiments_Results/ – generated performance plots (MAE, MSE, etc.).
• Model_comparison_result/ – final comparison graph of different models.
• requirements.txt – pinned Python dependencies.
• No separate frontend/backend; the notebook drives everything.
• Results folders act as storage for outputs from various pipeline stages.
README for Insights
• Dependencies highlight major libraries: Pandas/Polars/Dask for data; TensorFlow/Keras/scikit-learn for modeling; Plotly/Folium/Matplotlib for visualization.
• Emphasis on interactive prototyping and consistent environment via requirements.txt.
Main Components and Their Interactions
1. Data Source (CSV)
• Reads Data/bitcoin.csv via Pandas/Polars/DuckDB
2. Data Processing
• Uses Dask or Polars, NumPy, SymPy for any transformations
3. Model Training
• Uses TensorFlow/Keras or scikit-learn to fit forecasting models
4. Experiment Evaluation
• Calculates metrics (MAE, MSE, RMSE, etc.), saves plots in Experiments_Results
5. Visualization & Reporting
• Renders comparison in the notebook and saves final graph under Model_comparison_result
6. Dependency Management
• requirements.txt ensures reproducibility; runs under pip environment
1 → 2 → 3 → 4 → 5
Architectural Patterns and Design Principles
• Pipeline pattern: sequential data transformations and model stages.
• Reproducibility: explicit dependency management.