I compared both the list_object function of minio and the list_object function that I directly override(NAS class). At this time, only when the prefix is'metadata/test/', there is a problem that the two do not match. Be careful about this.
Add unittests for NAS
data saver and data loader
model saver and model loader
optimizer saver and optimizer loader
As one problem, we found a problem that unittest.skip cannot be called in the same class. Therefore, this stackoverflow answer solved this.
https://github.com/graykode/matorage/pull/29/commits/e43fd939bacbee935c11a2b4db0bed8c503928c5 : Unlike pytorch, tensorflow cannot upload a model twice. Therefore, the unittest function related to reload was removed. To make this possible, it is necessary to divide it into a train process using multiprocess and an eval process that reloads it after training. Previously, it was not a problem because you did not enter this function with unittest.skip, but when entering, a deadlock occurs in the join of the multiprocess.
21
I compared both the
list_object
function of minio and thelist_object
function that I directly override(NAS class). At this time, only when the prefix is'metadata/test/', there is a problem that the two do not match. Be careful about this.Add unittests for NAS
As one problem, we found a problem that
unittest.skip
cannot be called in the same class. Therefore, this stackoverflow answer solved this.