This repository contains a full pipeline for collecting, preprocessing, training, and forecasting cryptocurrency data using a transformer-based model called LiteFormer. The project integrates with the Polygon API to fetch historical crypto data, applies technical analysis indicators using the ta library, and builds a deep learning model with PyTorch to predict future close prices.

• Overview
• Features
• Installation
• Usage
• Code Structure
• Directory Structure
• Configuration
• Contributing
• License
• Contact

The project implements a complete workflow for cryptocurrency forecasting:

1. Data Collection:
   Fetches data from the Polygon API for given tickers and timeframes, saving raw data into CSV files.

2. Data Preprocessing:
   Computes technical indicators such as RSI, MACD, ATR, Bollinger Bands, and ADX, and prepares the data for modeling.

3. Model Training:
   Uses a transformer-based model (LiteFormer) with positional encoding for time series forecasting. Training includes features like gradient accumulation, learning rate scheduling, and early stopping.

4. Prediction & Evaluation:
   Generates predictions for the next time steps, computes error metrics (MAE, RMSE), and categorizes predictions.

5. Result Sharing:
   Optionally uploads the combined predictions CSV file to Oshi for easy sharing.

• Multi-Ticker Data Collection: Fetch historical data for multiple cryptocurrencies.
• Technical Indicator Calculation: Leverages the ta library for RSI, MACD, ATR, Bollinger Bands, and ADX.
• Transformer-Based Forecasting: Implements a custom transformer (LiteFormer) model using PyTorch.
• Custom Dataset & DataLoader: Prepares time-series windows from the data for model training.
• Training & Validation Pipeline: Uses gradient accumulation and adaptive learning rate scheduling.
• Prediction & Metrics: Evaluates model performance and saves predictions with additional classification (direction, percentage change).
• File Upload: Supports uploading results to Oshi for public access.

1. Clone the repository:

   git clone https://github.com/yourusername/crypto-forecasting.git
   cd crypto-forecasting