A computational study was conducted to investigate the mechanical behavior of a nanoporous gold (np-Au) structure by molecular dynamics (MD) simulations. In these MD simulations the mechanical load on the samples was applied in form of multi-axial stress-controlled boundary conditions that are systematically distributed over the entire stress space. In this way, the mechanical response of the np-Au structure under a wide range of mechanical loading conditions, including tension, compression and shear components in each load case, was obtained in form of stress-strain data. It is seen that the mechanical properties of np-Au are rather anisotropic in nature because the arrangement of the gold atoms within the nanoporous structure is inherited from a gold monocrystal. Furthermore, the porous structure exhibits a significant tension-compression asymmetry in its mechanical behavior and, in contrast to solid metals, the plastic deformation strongly depends on hydrostatic stress components. While there are well-established phenomenological flow rules that describe severe plastic anisotropy or others that are able to capture the influence of hydrostatic tension or compression, the combination of both phenotypes of plastic behavior has rarely been addressed in constitutive modeling. In the present work, it is demonstrated that a machine learning plasticity model trained by the data of the atomistic simulations captures the mechanical properties of the np-Au structures in both aspects with an accuracy of 96%. This was achieved by training a support vector classification (SVC) model to discriminate between elastic and plastic regions in a 12-dimensional feature space spanned by the independent components of stress and plastic strain tensors. Hence, such data-based constitutive models provide an efficient way to bridge the length scales from atomistic simulations to macroscopic materials modeling.