[train] remove default num_workers (#19518)

* [train] remove default num_workers

* fix tests

python/ray/train/callbacks/logging.py

@@ -24,7 +24,7 @@ class TrainingLogdirMixin:
             logdir_path = Path(logdir)
 
         if not logdir_path.is_dir():
-            raise ValueError(f"logdir '{logdir}' must be a directory.")
+            raise ValueError(f"logdir '{logdir_path}' must be a directory.")
 
         self._logdir_path = logdir_path
 

python/ray/train/examples/mlflow_fashion_mnist_example.py

@@ -6,7 +6,7 @@ from ray.train import Trainer
 from ray.train.examples.train_fashion_mnist_example import train_func
 
 
-def main(num_workers=1, use_gpu=False):
+def main(num_workers=2, use_gpu=False):
     mlflow.set_experiment("train_torch_fashion_mnist")
 
     trainer = Trainer(
@@ -38,7 +38,7 @@ if __name__ == "__main__":
         "--num-workers",
         "-n",
         type=int,
-        default=1,
+        default=2,
         help="Sets number of workers for training.")
     parser.add_argument(
         "--use-gpu",

python/ray/train/examples/train_fashion_mnist_example.py

@@ -120,7 +120,7 @@ def train_func(config: Dict):
     return loss_results
 
 
-def train_fashion_mnist(num_workers=1, use_gpu=False):
+def train_fashion_mnist(num_workers=2, use_gpu=False):
     trainer = Trainer(
         backend="torch", num_workers=num_workers, use_gpu=use_gpu)
     trainer.start()
@@ -131,7 +131,7 @@ def train_fashion_mnist(num_workers=1, use_gpu=False):
             "batch_size": 64,
             "epochs": 4
         },
-        callbacks=[JsonLoggerCallback("./train_results")])
+        callbacks=[JsonLoggerCallback()])
     trainer.shutdown()
     print(f"Loss results: {result}")
 
@@ -147,15 +147,13 @@ if __name__ == "__main__":
         "--num-workers",
         "-n",
         type=int,
-        default=1,
+        default=2,
         help="Sets number of workers for training.")
     parser.add_argument(
         "--use-gpu",
         action="store_true",
         default=False,
         help="Enables GPU training")
-    parser.add_argument(
-        "--tune", action="store_true", default=False, help="Tune training")
     parser.add_argument(
         "--smoke-test",
         action="store_true",

python/ray/train/tests/test_trainer.py

@@ -592,10 +592,10 @@ def test_horovod_torch_mnist_stateful(ray_start_2_cpus):
 
 def test_init_failure(ray_start_2_cpus):
     with pytest.raises(TypeError):
-        Trainer(5)
+        Trainer(5, num_workers=2)
 
     with pytest.raises(ValueError):
-        Trainer("invalid")
+        Trainer("invalid", num_workers=2)
 
 
 def test_start_failure(ray_start_2_cpus):

python/ray/train/tests/test_tune.py

@@ -97,7 +97,7 @@ def test_tune_error(ray_start_2_cpus):
     def train_func(config):
         raise RuntimeError("Error in training function!")
 
-    trainer = Trainer(TestConfig())
+    trainer = Trainer(TestConfig(), num_workers=1)
     TestTrainable = trainer.to_tune_trainable(train_func)
 
     with pytest.raises(TuneError):
@@ -110,7 +110,7 @@ def test_tune_checkpoint(ray_start_2_cpus):
             train.report(test=i)
         train.save_checkpoint(hello="world")
 
-    trainer = Trainer(TestConfig())
+    trainer = Trainer(TestConfig(), num_workers=1)
     TestTrainable = trainer.to_tune_trainable(train_func)
 
     [trial] = tune.run(TestTrainable).trials
@@ -133,7 +133,7 @@ def test_reuse_checkpoint(ray_start_2_cpus):
             train.save_checkpoint(iter=i)
             train.report(test=i, training_iteration=i)
 
-    trainer = Trainer(TestConfig())
+    trainer = Trainer(TestConfig(), num_workers=1)
     TestTrainable = trainer.to_tune_trainable(train_func)
 
     [trial] = tune.run(TestTrainable, config={"max_iter": 5}).trials
@@ -163,7 +163,7 @@ def test_retry(ray_start_2_cpus):
             train.save_checkpoint(iter=i)
             train.report(test=i, training_iteration=i)
 
-    trainer = Trainer(TestConfig())
+    trainer = Trainer(TestConfig(), num_workers=1)
     TestTrainable = trainer.to_tune_trainable(train_func)
 
     analysis = tune.run(TestTrainable, max_failures=3)

python/ray/train/trainer.py

@@ -64,8 +64,8 @@ class Trainer:
             a subclass of ``BackendConfig`` can be passed in.
             Supported ``str`` values: {"torch", "tensorflow", "horovod"}.
         num_workers (int): The number of workers (Ray actors) to launch.
-            Defaults to 1. Each worker will reserve 1 CPU by default. The
-            number of CPUs reserved by each worker can be overridden with the
+            Each worker will reserve 1 CPU by default. The number of CPUs
+            reserved by each worker can be overridden with the
             ``resources_per_worker`` argument.
         use_gpu (bool): If True, training will be done on GPUs (1 per
             worker). Defaults to False. The number of GPUs reserved by each
@@ -85,7 +85,7 @@ class Trainer:
     def __init__(
             self,
             backend: Union[str, BackendConfig],
-            num_workers: int = 1,
+            num_workers: int,
             use_gpu: bool = False,
             resources_per_worker: Optional[Dict[str, float]] = None,
             logdir: Optional[str] = None,