Closed jonas-eschle closed 3 years ago
We also recently moved to the tnp interface in zfit (https://github.com/zfit/zfit/pull/310) which seemed to mostly work, but it's tricky as it is still experimental and they mean it. For example, while before a TF.ndarray was returned, it is now a TF.Tensor again; however a new numpy behavior switch was introduced (https://www.tensorflow.org/api_docs/python/tf/experimental/numpy/experimental_enable_numpy_behavior)
But overall it's clear that we want to go in this direction to have not only numpy but also jax etc in the future as a possible backend
We also recently moved to the tnp interface in zfit (https://github.com/zfit/zfit/pull/310) which seemed to mostly work, but it's tricky as it is still experimental and they mean it. For example, while before a TF.ndarray was returned, it is now a TF.Tensor again; however a new numpy behavior switch was introduced (https://www.tensorflow.org/api_docs/python/tf/experimental/numpy/experimental_enable_numpy_behavior)
But overall it's clear that we want to go in this direction to have not only numpy but also jax etc in the future as a possible backend