stat386-final 0.1.8

Statistical analysis of video game sales data with preprocessing, visualization, and predictive modeling

Sales Forecast ML Package

A Python package for reading, preprocessing, visualizing, and modeling sales data. It provides utilities to:

• Load tabular sales data from CSV (read.read_data)
• Clean and engineer features (preprocess.process_data, preprocess.prepare_data)
• Visualize sales distributions by genre or platform (viz.print_genre_distribution, viz.print_platform_distribution)
• Train and tune a Random Forest model and generate predictions (model.rf_fit, model.predict)

---

Overview

This package is designed to help you work with video game sales and activity data. It includes tools for reading raw CSV files, preprocessing and feature engineering, visualizing distributions, and building predictive models using Random Forest.

---

Features

• CSV Reader: Reads a CSV file and applies basic type cleaning for year, drops an extraneous index column.
• Data Processing: Aggregates per Name and combines sales/metadata, removes duplicates.
• Feature Preparation: Multi-label binarization for Platform and Genre; concatenates with numeric predictors, drops missing Year.
• Visualization:
  • print_genre_distribution(sales, genre, area): Histogram of sales for a given genre and region.
  • print_platform_distribution(sales, platform, area): Histogram of sales for a given platform and region.
• Modeling: Grid search over Random Forest hyperparameters; evaluation prints R², RMSE (log scale), and top feature importances.
• Prediction: Scales inputs consistently and returns predictions on the original scale via expm1.

---

Included Datasets

The package includes two sample datasets in the data/ directory for quick experimentation:

• data/vgsales.csv: Raw sales data with columns like Name, Platform, Year, Genre, Publisher, regional sales, and Global_Sales.
• data/game_data.csv: Steam activity metrics (all_time_peak, last_30_day_avg) for integration with sales data aggrigated with the vgsales dataset.

Use get_data_path(filename) from read.py to access these files programmatically.

---

Installation

pip install -U pip
pip install -U scikit-learn pandas numpy matplotlib seaborn
# If packaged for PyPI:
# pip install your-package-name

---

Quickstart

from your_package import (
    read_data, process_data, prepare_data,
    print_genre_distribution, print_platform_distribution,
    rf_fit, predict
)

# 1) Read
data = read_data("data/sales.csv")

# 2) Process & Prepare
sales_combined = process_data(data)
final_df = prepare_data(sales_combined)

# 3) Visualize
print_genre_distribution(data, genre="Action", area="Global_Sales")
print_platform_distribution(data, platform="PS4", area="Global_Sales")

# 4) Train model
best_model = rf_fit(final_df, area="Global_Sales")

# 5) Predict
new_data = final_df.drop(columns=["Global_Sales"]).iloc[:5]
preds = predict(best_model, area="Global_Sales", new_data=new_data)
print(preds)

---

API Reference

Top-Level Functions

• read_data(file_path)
• process_data(df)
• prepare_data(df)
• print_genre_distribution(sales, genre, area)
• print_platform_distribution(sales, platform, area)
• rf_fit(final_df, area)
• predict(model, area, new_data)

---

Dependencies

• pandas
• numpy
• scikit-learn
• matplotlib
• seaborn

Contributing

Issues and PRs are welcome. Please include reproducible examples and data schema.