Extensive empirical evidence reveals that, for a wide range of different\nlearning methods and datasets, the risk curve exhibits a double-descent (DD)\ntrend as a function of the model size. In a recent paper\n[Zeyu,Kammoun,Thrampoulidis,2019] the authors studied binary linear\nclassification models and showed that the test error of gradient descent (GD)\nwith logistic loss undergoes a DD. In this paper, we complement these results\nby extending them to GD with square loss. We show that the DD phenomenon\npersists, but we also identify several differences compared to logistic loss.\nThis emphasizes that crucial features of DD curves (such as their transition\nthreshold and global minima) depend both on the training data and on the\nlearning algorithm. We further study the dependence of DD curves on the size of\nthe training set. Similar to our earlier work, our results are analytic: we\nplot the DD curves by first deriving sharp asymptotics for the test error under\nGaussian features. Albeit simple, the models permit a principled study of DD\nfeatures, the outcomes of which theoretically corroborate related empirical\nfindings occurring in more complex learning tasks.\n