Sep 5, 2018 · 1 min read
I’ve just dipped my toe into linear regression in TensorFlow(TM), and I was surprised by the absence of the usual diagnostics. MSA and MAE are optionally available, but I could see nothing in the documentation on Intercept, coefficients, t-value F-statistic, R-squared Adjusted R-squared, p-value, residuals, null deviance and the less common measures. I assume it’s possible to pass a model specification from scipy.stats and retrieve the diagnostics out of the TM stream. Have you seen that done? Thanks.
