I want to implement a custom metric (pearson correlation) as defined here in Keras. I get a batch (32) of predictions and labels. I use a neural network to predict 10 values. So my input is 32x10 predictions and 32x10 labels.

I would like to know how I get a single correlation value for batches of predictions/labels. Should I calculate the correlation for each prediction/label sample in the batch individually and then take the mean over the 32 samples or should I treat the batch as a "population" and use the covariance between prediction and label and divide it by the standard deviations?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.