Torch-metrics: model evaluation metrics for PyTorch

As summarized in this issue, Pytorch does not have a built-in libary torch.metrics for model evaluation metrics. This python library serves as a custom library to provide common evaluation metrics in Pytorch, similar to tf.keras.metrics. This is similar to the metrics library in PyTorch Lightning.

Usage

• pip install --upgrade torch-metrics or
• git clone https://github.com/enochkan/torch-metrics.git

from torch_metrics import Accuracy

## define metric ##
metric = Accuracy(from_logits=False)
y_pred = torch.tensor([1, 2, 3, 4])
y_true = torch.tensor([0, 2, 3, 4])

print(metric(y_pred, y_true))

from torch_metrics import Accuracy

## define metric ##
metric = Accuracy()
y_pred = torch.tensor([[0.2, 0.6, 0.1, 0.05, 0.05],
                       [0.2, 0.1, 0.6, 0.05, 0.05],
                       [0.2, 0.05, 0.1, 0.6, 0.05],
                       [0.2, 0.05, 0.05, 0.05, 0.65]])
y_true = torch.tensor([0, 2, 3, 4])

print(metric(y_pred, y_true))

Implementation

Metrics from tf.keras.metrics and other metrics that are already implemented vs to-do

• MeanSquaredError class
• RootMeanSquaredError class
• MeanAbsoluteError class
• Precision class
• Recall class
• MeanIoU class
• DSC class (Dice Similarity Coefficient)
• F1Score class
• RSquared class
• Hinge class
• SquaredHinge class
• LogCoshError class
• Accuracy class
• KLDivergence class
• CosineSimilarity class
• AUC class
• BinaryCrossEntropy class
• CategoricalCrossEntropy class
• SparseCategoricalCrossentropy class

Please raise issues or feature requests here.