EXPLICIT FEATURE SPACES FOR LEARNING WITH GRAPHS In this project, several graph kernels are evaluated in terms of classification accuracy and runtime. Specifically, one implicit and five explicit embedding methods are assessed. The random walk kernel is the implicit embedding, while the explicit methods comprise the Weisfeiler-Lehman subtree kernel, the neighborhood hash kernel (in three variants), the graphlet kernel (in two variants), the shortest path kernel and the Eigen graph kernel. The respective implementations can be found in the folder "code/embeddings". The main module is "code/eval_embeddings.py". In this module, the embeddings can be specified that are to be evaluated. Furthermore, the datasets can be determined, on which the embeddings will run. See the comments within "code/eval_embeddings.py" for more information. The classification accuracies and runtimes were evaluated on the following eight datasets: MUTAG, PTC(MR), ENZYMES, DD, NCI1, NCI109, FLASH CFG, and ANDROID FCG. The datasets are contained in the folder "datasets". The results of the experiments presented in the master's thesis can be found in the folder "results_final".