Open krzentner opened 5 years ago
@krzentner is this still an issue? It appears Box
is the only class that takes a dtype in the constructor and there is code to support np.float32
and np.uint8
. However, all of the tensorflow converters assume a dtype of tf.uint8
. Let me know if those should change to support tf.float32
as well.
This is still an issue, although not a very high priority one. akro.Tuple
actually uses np.core.numerictypes.find_common_type
to find the common dtype
. It would be good to preserve that behavior, and spread it to akro.Dict
.
Currently, all of the spaces assume that the datatype is np.float32. However, some environments use e.g. np.uint8. We should support that.
See garage PR 455.