What is a good cross-entropy loss value?

Lower is better. 0.0 means perfect predictions. For binary classification, random guessing gives ~0.693. For N-class classification, random guessing gives ln(N). Values well below the random baseline indicate the model has learned something useful.

Why must predicted probabilities sum to 1 in categorical mode?

Categorical cross-entropy assumes a proper probability distribution over classes (e.g., softmax output). If probabilities do not sum to 1, the formula is technically undefined. The calculator warns you but still computes a result.

What is the difference between log loss and cross-entropy?

For binary classification they are the same thing. "Log loss" is the common scikit-learn name; "binary cross-entropy" is the deep-learning name. Both use the same formula.

Can I compute the average loss over multiple samples?

This calculator computes loss for a single sample. To get the mean over a dataset, sum the per-sample losses and divide by the number of samples.

What is the range of cross-entropy loss?

Cross-entropy loss ranges from 0 to infinity. A perfect prediction gives 0, while a confident wrong prediction pushes the value toward infinity. There is no fixed upper bound, so compare against the random-guess baseline of ln(N) for N classes.

Cross-Entropy Loss Calculator

Name: Cross-Entropy Loss Calculator
Availability: InStock
Author: Nham Vu

Calculate binary or categorical cross-entropy loss between predicted probabilities and true labels for machine learning model evaluation.

Mode

True label

Predicted probability (0–1)

Result

Enter values and click Calculate

Reference — Common Values

Loss	Meaning

Summary

Calculate binary or categorical cross-entropy loss between predicted probabilities and true labels for machine learning model evaluation.

How it works

Select Binary mode for two-class problems (true label 0 or 1, predicted probability 0–1).
Select Categorical mode for multi-class problems — enter comma-separated one-hot true labels and predicted probabilities.
Binary formula: H = -[y × log(p) + (1-y) × log(1-p)].
Categorical formula: H = -Σ(y_i × log(p_i)) over all classes.
The result shows the loss value, entropy interpretation, and a reference table for context.

Use cases

Verify that a neural network loss value matches manual calculation during debugging.
Understand what a reported cross-entropy loss means in practice (good vs. random vs. bad).
Teach or learn the difference between binary and categorical cross-entropy.
Quickly check whether predicted probabilities for a single sample are reasonable.

Frequently Asked Questions

Last updated: 2026-06-11 · Reviewed by Nham Vu