In the era of massive all-sky surveys, calculating spectroscopic redshifts (Spec-Z) is time-consuming and costly. Instead, broadband photometry may be used as a proxy for spectrographic measurements. We train an off-the-shelf classifier to estimate photometric redshifts (Photo-Z) from SDSS Quasar imaging data. We outline the technique of cross-validation to reduce bias and variance in the model and improve calibration. Reasons for so-called catastrophic prediction failures are discussed and popular packages such as scikit-learn and AstroML examined. The modern Deep Learning package 'fastai' is introduced along with a survey of Neural Net (NN) approaches. Despite Machine Learning being something of a dark art, surprisingly useful results may be extracted using modern software packages.