Metrics Report

MetricsReport is a Python package that generates classification and regression metrics report for machine learning models.

Features

• AutoDetect the type of task
• Save report in .html and .md format
• Has several plotting functions

Installation

You can install MetricsReport using pip:

pip install metricsreport

Usage

from metricsreport import MetricsReport  

# sample classification data 
y_true = [1, 0, 0, 1, 0, 1, 0, 1] 
y_pred = [0.8, 0.3, 0.1, 0.9, 0.4, 0.7, 0.2, 0.6]  

# generate report 
report = MetricsReport(y_true, y_pred, threshold=0.5)  

# print all metrics 
print(report.metrics)  

# plot ROC curve 
report.plot_roc_curve()

# saved MetricsReport (html) in folder: report_metrics
report.save_report()

More examples in the folder ./examples:

Constructor

MetricsReport(y_true, y_pred, threshold: float = 0.5)

• y_true : list
  • A list of true target values.
• y_pred : list
  • A list of predicted target values.
• threshold : float
  • Threshold for generating binary classification metrics. Default is 0.5.

Plots

following methods can be used to generate plots:

• plot_roc_curve(): Generates a ROC curve plot.
• plot_all_count_metrics(): Generates a count metrics plot.
• plot_precision_recall_curve(): Generates a precision-recall curve plot.
• plot_confusion_matrix(): Generates a confusion matrix plot.
• plot_class_distribution(): Generates a class distribution plot.
• plot_class_hist(): Generates a class histogram plot.
• plot_calibration_curve(): Generates a calibration curve plot.
• plot_lift_curve(): Generates a lift curve plot.
• plot_cumulative_gain(): Generates a cumulative gain curve plot.

Dependencies

• numpy
• pandas
• matplotlib
• scikit-learn
• scikit-plot

License

This project is licensed under the MIT License.