Brent A. Thorne

There appears to have been an API change to HistogramWrapper. It's Return type is now a capsa.RiskTensor containing both the predicted y_hat (tf.Tensor) and the bias value for each input...

Here is a messy workaround for this data type casting issue... ` for step, data in enumerate(train_loader): x,y = data xx = [tf.convert_to_tensor(a) for a in x] yy = [tf.convert_to_tensor(b)...