In traditional machine learning models, we use predict and it gives the point that always lies on line of best fit. Taking example of regression, but in reality the point could live in variance of line best fit.

Since the output of tensorflow probability model having distribution lambda in output gives the distribution that represents the data coming from previous layers.

So lets say I am customer and devs have used `model.predict(x)`

and it gives me different values for same input, how can I trust whether single output is correct?