Decision Tree Analysis - WiFi Localization

Requirements

• Python 3.7+
• NumPy
• Matplotlib

Installation

pip install numpy matplotlib

Usage

1. Complete Analysis Pipeline (Default)

Runs the full decision tree analysis including tree building, visualization, and cross-validation on both clean and noisy datasets.

python3 run.py
python3 run.py --random_seed 42

What it does:

• Loads clean dataset and splits into train/test (80/20)
• Builds decision tree on training data
• Generates tree visualization
• Performs 10-fold cross-validation on clean dataset
• Performs 10-fold cross-validation on noisy dataset
• Saves evaluation plots and metrics

Options:

• --random_seed INT: Set random seed for reproducible results (default: -1, 42 can be used to exactly replicate results in report)

Output Files: Output files are stored in location ./CW1/docs/figures

• tree_clean.png: Decision tree visualization
• decision_tree_evaluation_clean.png: Clean data evaluation metrics
• decision_tree_evaluation_noisy.png: Noisy data evaluation metrics

2. Build and Evaluate Single Decision Tree

Builds a single decision tree on specified dataset and evaluates its performance.

python3 run.py build_tree
python3 run.py build_tree --data_path wifi_db/noisy_dataset.txt --visualize
python3 run.py build_tree --train_split 0.7 --random_seed 123

What it does:

• Loads specified dataset
• Splits data according to train_split ratio
• Builds decision tree on training data
• Evaluates tree on test data
• Optionally generates tree visualization
• Prints accuracy and performance metrics

Options:

• --data_path PATH: Dataset file path (Data must be present in location ./CW1/data/ and path supplied must be relative to this location)
• --train_split FLOAT: Training data ratio, 0.1 - 0.9 (default: 0.8)
• --random_seed INT: Random seed for reproducibility (default: -1)
• --visualize: Generate and save tree visualization

Output:

• Console: Accuracy, confusion matrix, precision, recall, F1-scores
• Optional: Tree visualization PNG file

3. K-Fold Cross-Validation

Performs k-fold cross-validation on specified dataset and reports averaged metrics.

python3 run.py cross_validate
python3 run.py cross_validate --k 5 --data_path wifi_db/noisy_dataset.txt
python3 run.py cross_validate --k 10 --random_seed 42

What it does:

• Loads specified dataset
• Performs k-fold cross-validation
• Trains decision tree on each training fold
• Evaluates on each test fold
• Computes averaged metrics across all folds
• Generates evaluation visualization

Options:

• --k INT: Number of cross-validation folds, minimum 2 (default: 10)
• --data_path PATH: Dataset file path (default: wifi_db/clean_dataset.txt)
• --random_seed INT: Random seed for reproducibility (default: -1)

Output:

• Console: Average accuracy, confusion matrix, precision, recall, F1-scores
• PNG file: Cross-validation evaluation plots

4. Help and Usage Information

python3 run.py --help
python3 run.py -h

What it shows:

• Detailed command descriptions
• All available options and their defaults
• Usage examples for each command
• Parameter validation information

Dataset Paths

• Clean dataset: wifi_db/clean_dataset.txt
• Noisy dataset: wifi_db/noisy_dataset.txt

Output Files Location

All generated files are saved in the docs/figures/ directory:

• Tree visualizations: tree_*.png
• Evaluation plots: decision_tree_evaluation_*.png

Validation and Error Handling

• Training split must be between 0.1 and 0.9
• Cross-validation folds must be ≥ 2
• Warns for computationally expensive operations (k > 20)
• Provides helpful error messages and suggestions