Only do numpy seeding if numpy seed is not fixed, fix regression intr…

…oduced by 9f9e12f

kge/job/train.py

@@ -21,6 +21,7 @@
 SLOTS = [0, 1, 2]
 S, P, O = SLOTS
 
+
 def _worker_init_fn(worker_num):
  # ensure that NumPy uses different seeds at each worker
  np.random.seed()
@@ -595,7 +596,9 @@ def _prepare(self):
  shuffle=True,
  batch_size=self.batch_size,
  num_workers=self.config.get("train.num_workers"),
- worker_init_fn=_worker_init_fn,
+ worker_init_fn=_worker_init_fn
+ if self.config.get("random_seed.numpy") == -1
+ else None,
  pin_memory=self.config.get("train.pin_memory"),
  )
 
@@ -803,7 +806,9 @@ def _prepare(self):
  shuffle=True,
  batch_size=self.batch_size,
  num_workers=self.config.get("train.num_workers"),
- worker_init_fn=_worker_init_fn,
+ worker_init_fn=_worker_init_fn
+ if self.config.get("random_seed.numpy") == -1
+ else None,
  pin_memory=self.config.get("train.pin_memory"),
  )
 
@@ -1038,7 +1043,9 @@ def _prepare(self):
  shuffle=True,
  batch_size=self.batch_size,
  num_workers=self.config.get("train.num_workers"),
- worker_init_fn=_worker_init_fn,
+ worker_init_fn=_worker_init_fn
+ if self.config.get("random_seed.numpy") == -1
+ else None,
  pin_memory=self.config.get("train.pin_memory"),
  )