How vaggala/movie-revenue-predictor works
Project type and purpose
• It reads structured CSV data, applies preprocessing and feature engineering (including NLP on the “overview” text), trains a Random Forest regression model, evaluates performance, and produces visualizations.
File structure
• Top level:
• data/ – raw CSV inputs (tmdb_5000_movies.csv, tmdb_5000_credits.csv)
• notebooks/ – exploratory Jupyter notebooks
• src/ – core pipeline modules (load_data.py, preprocessing.py, train_model.py, evaluate_model.py)
• main.py – orchestrator script that ties all stages together
• visualizations/ – output plots (Actual vs. Predicted, etc.)
• Configs & metadata: README.md, .gitignore, .gitattributes
README for additional insights
• Tech stack: scikit-learn, pandas, numpy, matplotlib, Hugging Face Transformers
• Model: Random Forest Regression, plus sentiment analysis on the “overview” field
• Metrics: R², MAE, RMSE
• Future improvements hint at model persistence (joblib), hyperparameter tuning, more features
Main components
a. Data Ingestion Layer
– Source: CSV files in data/
– Module: src/load_data.py
b. Preprocessing & Feature Engineering
– Structured: budget, cast size, runtime, genres, popularity, votes
– Unstructured/NLP: text embedding & sentiment analysis via Hugging Face Transformers (src/preprocessing.py)
c. Model Training
– Algorithm: Random Forest Regressor (src/train_model.py)
d. Model Evaluation
– Metrics calculation (R², MAE, RMSE) in src/evaluate_model.py
e. Visualization
– Plot generation in visualizations/ (Actual vs. Predicted)
f. Orchestration
Relationships and interactions
• CSV files → Data Loader → preprocessed Data
• Preprocessed Data → Feature Engineering → Feature Set
• Feature Set → Model Training → Trained Model
• Trained Model + Test Set → Evaluation → Metrics
• Metrics + Predictions → Visualization → Plots
• External service: Hugging Face Transformers called during NLP feature step
Architectural patterns & design principles
• Modular pipeline: each Python module has a single responsibility
• Clear separation of concerns: data ingestion, transformation, modeling, evaluation
• Reproducibility: orchestration via a single entrypoint (main.py)
• Extensibility: easy to swap in new models (e.g., XGBoost), add features, or persist models with joblib
Technologies & frameworks to annotate
• Python 3.x
• pandas / numpy (data manipulation)
• scikit-learn (Random Forest)
• transformers (Hugging Face)
• matplotlib (visualization)
• Jupyter notebooks (exploration)