Evaluating model performance with torch-metrics

PyTorch is a powerful open-source machine learning library written in Python. Unlike Keras’s tf.keras.metrics, however, PyTorch does not have an out-of-the-box library for model evaluation metrics as illustrated in this github issue.

Photo by Aziz Acharki on Unsplash

torch-metrics is a library written for PyTorch model evaluation. To install torch-metrics , simply run pip install --upgrade torch-metrics and the latest version will be installed.

Available Metrics

At the time of writing, the following metrics are available:

• Accuracy
• R-squared
• Mean Absolute Error (MAE)
• Mean Squared Error (MSE)
• Root Mean Squared Error (RMSE)
• Precision
• Recall
• F1
• Mean IoU (Intersection over Union)
• Dice Similarity Coefficient (DSC)
• Hinge
• Structural Similarity (SSIM)

Usage

The goal is to eventually implement all the evaluation metrics available in the Keras metrics API. Below is an example of using torch-metrics to evaluate two PyTorch tensors.

Contributions are welcome. Visit this github repo for source code and latest updates.