Several strategies are proposed to parallelize graph neural network (GNN) training over multiple GPUs. These strategies are logically equivalent (i.e., produce the same model) but partition data and assign computation to the GPUs in different ways. We observe that there is no consistent winner (i.e., with the shortest running time), and the optimal strategy depends on the graph dataset, GNN model, training algorithm, and hardware configurations. As such, we design the APT system to automatically select efficient parallelization strategies for GNN training tasks. In particular, we analyze the trade-offs of the strategies and design simple yet effective cost models to compare their execution time and facilitate strategy selection. Moreover, we also propose a general abstraction of the strategies, which allows to implement a unified execution engine that can be configured to run different strategies. Our experiments show that APT runs the optimal strategy across various task configurations, and the training time can be reduced by over 2x compared with always using a single strategy. APT is open-source anonymously at https://anonymous.4open.science/r/APT-1CAB.