metadata

title: Top5 Error Rate
emoji: 📈
colorFrom: yellow
colorTo: blue
sdk: gradio
sdk_version: 5.24.0
app_file: app.py
pinned: false
tags:
  - evaluate
  - metric

Metric Card for Top-5 error rate

Metric Description

The "top-5 error" is the percentage of times that the target label does not appear among the 5 highest-probability predictions. It can be computed with: Top-5 Error Rate = 1 - Top-5 Accuracy or equivalently: Top-5 Error Rate = (Number of incorrect top-5 predictions) / (Total number of cases processed) Where:

Top-5 Accuracy: The proportion of cases where the true label is among the model's top 5 predicted classes.
Incorrect top-5 prediction: The true label is not in the top 5 predicted classes (ranked by probability).

How to Use

At minimum, this metric requires predictions and references as inputs.

accuracy_metric = evaluate.load("Aye10032/top5_error_rate")

labels: torch.Tensor = batch_data['labels']
train_output = model(datas)

results = accuracy_metric.compute(references=train_output.cpu(), predictions=labels)
print(results)

output is

{'top5_error_rate': ..., 'accuracy': ...}