How ranakroychowdhury/multimodal-deep-learning-based-stock-price-prediction works
Project type and purpose
• This is a data-science / machine-learning pipeline project that uses deep learning (LSTM) to predict stock prices from historical price data and news headlines.
• Core objectives:
• Ingest and preprocess two data sources (stock prices, news headlines)
• Train an LSTM neural network using Keras
• Tune hyperparameters
File structure
• Top-level scripts:
• data.py Data loading & preprocessing
• train.py Training the LSTM model
• tune.py Hyperparameter search
• test.py Model evaluation & test runs
• script.bat, script.sh Command-line wrappers
• Directories & support files:
• Data/All_new.csv Combined stock + news dataset
• Lib/config.py Global configuration (paths, hyperparameter defaults)
• Report.docx, Results.docx Documentation & outputs
• No frontend or deployment code—this is an offline training pipeline
README for additional insights
• Data sources: Kaggle (news), Yahoo Finance (stock prices)
• Libraries & frameworks: Python, Keras for LSTM, likely pandas/numpy for preprocessing
• Outputs: performance plots, optimal parameters, predicted vs. actual price graphs
a. Main components:
1. Data Sources
– Kaggle headlines CSV
– Yahoo Finance historical prices
2. Data Ingestion & Preprocessing (data.py)
– Parsing CSVs, cleaning text, feature engineering (embeddings, scaling)
3. Configuration Module (Lib/config.py)
– Centralized hyperparameter and path settings
4. Model Training (train.py)
– Defines LSTM model in Keras
– Trains on preprocessed features
5. Hyperparameter Tuning (tune.py)
– Grid or random search over LSTM parameters
6. Evaluation & Testing (test.py)
– Computes metrics (MSE, MAE), holds out test set
7. Visualization & Reporting
– Plot predicted vs. actual, parameter performance curves
– Exports Results.docx, Report.docx
8. Execution Scripts (script.sh / script.bat)
– Orchestrate end-to-end runs
b. Relationships and data flow
– Scripts sit at the top, invoking each stage in sequence
c. Architectural patterns & design principles
– Pipeline pattern: distinct stages for ETL, training, tuning, evaluation
– Separation of concerns: config, data logic, model logic, reporting
– Modularity: each script focuses on one responsibility
d. Key technologies, frameworks, libraries
– Python ecosystem: pandas, numpy, scikit-learn (scaling, metrics)
– Deep learning: Keras (TensorFlow backend)
– Configuration: custom config.py
– Orchestration: shell/batch scripts