Length of training path affects results for reflected triangle.

(A) Training only on first view (as experienced from centre of the arena). (B) Training up until halfway to the feeder location ( N train = 15). (C) Training on the whole path ( N train = 30). (i) The training shape as it is experienced from the last training location. (ii) The test shape. (iii) The novelty sum output of the trained model after testing. We display the curves for 5 different parameter settings (crossover, overlap = [0,], [0.1, ], [0.2, ], [0.3, ], [0.4, ]) (iv) The novelty difference output of the model. (v) The estimated distributions (pooled) for all three modes of training. (TIF)