TF 2.0 is out! Get hands-on practice at TF World, Oct 28-31. Use code TF20 for 20% off select passes. Register now


View source on GitHub

Compute Brier score for a probabilistic prediction.


The [Brier score][1] is a loss function for probabilistic predictions over a number of discrete outcomes. For a probability vector p and a realized outcome k the Brier score is sum_i p[i]*p[i] - 2*p[k]. Smaller values are better in terms of prediction quality. The Brier score can be negative.

Compared to the cross entropy (aka logarithmic scoring rule) the Brier score does not strongly penalize events which are deemed unlikely but do occur, see [2]. The Brier score is a strictly proper scoring rule and therefore yields consistent probabilistic predictions.


[1]: G.W. Brier. Verification of forecasts expressed in terms of probability. Monthley Weather Review, 1950. [2]: Tilmann Gneiting, Adrian E. Raftery. Strictly Proper Scoring Rules, Prediction, and Estimation. Journal of the American Statistical Association, Vol. 102, 2007.


  • labels: Tensor, (N1, ..., Nk), with tf.int32 or tf.int64 elements containing ground truth class labels in the range [0, num_classes].
  • logits: Tensor, (N1, ..., Nk, num_classes), with logits for each example.
  • name: Python str name prefixed to Ops created by this function.


  • brier_score: Tensor, (N1, ..., Nk), containint elementwise Brier scores; caller should reduce_mean() over examples in a dataset.