Rename log_save_interval, row_log_interval

CHANGELOG.md

@@ -31,6 +31,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Changed
 
+- Change Trainer arguments `row_log_interval` and `log_save_interval` to
+    `log_every_n_steps` and `flush_logs_every_n_steps`, respectively.
+
 - Changed `LearningRateLogger` to `LearningRateMonitor` ([#3251](https://github.com/PyTorchLightning/pytorch-lightning/pull/3251))
 
 - Used `fsspec` instead of `gfile` for all IO ([#3320](https://github.com/PyTorchLightning/pytorch-lightning/pull/3320))

pytorch_lightning/trainer/__init__.py

@@ -574,15 +574,15 @@ def on_train_end(self, trainer, pl_module):
 
 .. note:: Might slow performance because it uses the output of nvidia-smi.
 
-log_save_interval
-^^^^^^^^^^^^^^^^^
+flush_logs_every_n_steps
+^^^^^^^^^^^^^^^^^^^^^^^^
 
 Writes logs to disk this often.
 
 .. testcode::
 
     # default used by the Trainer
-    trainer = Trainer(log_save_interval=100)
+    trainer = Trainer(flush_logs_every_n_steps=100)
 
 See Also:
     - :ref:`Experiment Reporting <experiment_reporting>`
@@ -909,15 +909,15 @@ def on_train_end(self, trainer, pl_module):
     # resume from a specific checkpoint
     trainer = Trainer(resume_from_checkpoint='some/path/to/my_checkpoint.ckpt')
 
-row_log_interval
-^^^^^^^^^^^^^^^^
+log_every_n_steps
+^^^^^^^^^^^^^^^^^
 
 How often to add logging rows (does not write to disk)
 
 .. testcode::
 
     # default used by the Trainer
-    trainer = Trainer(row_log_interval=50)
+    trainer = Trainer(log_every_n_steps=50)
 
 See Also:
     - :ref:`Experiment Reporting <experiment_reporting>`

pytorch_lightning/trainer/connectors/logger_connector.py

@@ -33,11 +33,11 @@ def __init__(self, trainer):
         self.progress_bar_metrics = {}
         self.eval_loop_results = []
 
-    def on_trainer_init(self, logger, log_save_interval, row_log_interval):
+    def on_trainer_init(self, logger, flush_logs_every_n_steps, log_every_n_steps):
         # logging
         self.configure_logger(logger)
-        self.trainer.log_save_interval = log_save_interval
-        self.trainer.row_log_interval = row_log_interval
+        self.trainer.flush_logs_every_n_steps = flush_logs_every_n_steps
+        self.trainer.log_every_n_steps = log_every_n_steps
 
     def configure_logger(self, logger):
         if logger is True:
@@ -470,7 +470,7 @@ def __gather_result_across_time_and_optimizers(self, epoch_output):
     def log_train_step_metrics(self, batch_output):
         # when metrics should be logged
         should_log_metrics = (
-            (self.trainer.global_step + 1) % self.trainer.row_log_interval == 0 or self.trainer.should_stop
+            (self.trainer.global_step + 1) % self.trainer.log_every_n_steps == 0 or self.trainer.should_stop
         )
         if should_log_metrics or self.trainer.fast_dev_run:
             # logs user requested information to logger

pytorch_lightning/trainer/trainer.py

@@ -107,8 +107,10 @@ def __init__(
         limit_val_batches: Union[int, float] = 1.0,
         limit_test_batches: Union[int, float] = 1.0,
         val_check_interval: Union[int, float] = 1.0,
-        log_save_interval: int = 100,
-        row_log_interval: int = 50,
+        log_save_interval: Optional[int] = None,  # backward compatible, todo: remove
+        row_log_interval: Optional[int] = None,  # backward compatible, todo: remove
+        flush_logs_every_n_steps: int = 100,
+        log_every_n_steps: int = 50,
         distributed_backend: Optional[str] = None,
         sync_batchnorm: bool = False,
         precision: int = 32,
@@ -177,6 +179,8 @@ def __init__(
 
             fast_dev_run: runs 1 batch of train, test and val to find any bugs (ie: a sort of unit test).
 
+            flush_logs_every_n_steps: How often to flush logs to disk (defaults to every 100 steps).
+
             gpus: number of gpus to train on (int) or which GPUs to train on (list or str) applied per node
 
             gradient_clip_val: 0 means don't clip.
@@ -191,7 +195,7 @@ def __init__(
 
             log_gpu_memory: None, 'min_max', 'all'. Might slow performance
 
-            log_save_interval: Writes logs to disk this often
+            log_every_n_steps: How often to log within steps (defaults to every 50 steps).
 
             prepare_data_per_node: If True, each LOCAL_RANK=0 will call prepare data.
                 Otherwise only NODE_RANK=0, LOCAL_RANK=0 will prepare data
@@ -230,8 +234,6 @@ def __init__(
             resume_from_checkpoint: To resume training from a specific checkpoint pass in the path here.
                 This can be a URL.
 
-            row_log_interval: How often to add logging rows (does not write to disk)
-
             sync_batchnorm: Synchronize batch norm layers between process groups/whole world.
 
             terminate_on_nan: If set to True, will terminate training (by raising a `ValueError`) at the
@@ -257,6 +259,15 @@ def __init__(
         """
         super().__init__()
 
+        # deprecation warnings
+        if row_log_interval is not None:
+            warnings.warn("row_log_interval is deprecated, use log_every_n_steps instead", DeprecationWarning)
+            log_every_n_steps = row_log_interval
+
+        if log_save_interval is not None:
+            warnings.warn("log_save_interval is deprecated, use flush_logs_every_n_steps instead", DeprecationWarning)
+            flush_logs_every_n_steps = log_save_interval
+
         # init connectors
         self.dev_debugger = InternalDebugger(self)
         self.config_validator = ConfigValidator(self)
@@ -291,7 +302,7 @@ def __init__(
             process_position,
             default_root_dir,
             weights_save_path,
-            resume_from_checkpoint
+            resume_from_checkpoint,
         )
 
         # hook
@@ -302,18 +313,12 @@ def __init__(
 
         # init data flags
         self.data_connector.on_trainer_init(
-            check_val_every_n_epoch,
-            reload_dataloaders_every_epoch,
-            prepare_data_per_node
+            check_val_every_n_epoch, reload_dataloaders_every_epoch, prepare_data_per_node
         )
 
         # init training tricks
         self.training_tricks_connector.on_trainer_init(
-            gradient_clip_val,
-            track_grad_norm,
-            accumulate_grad_batches,
-            truncated_bptt_steps,
-            terminate_on_nan
+            gradient_clip_val, track_grad_norm, accumulate_grad_batches, truncated_bptt_steps, terminate_on_nan
         )
 
         # init accelerator related flags
@@ -328,7 +333,7 @@ def __init__(
             sync_batchnorm,
             benchmark,
             replace_sampler_ddp,
-            deterministic
+            deterministic,
         )
 
         # init train loop related flags
@@ -342,7 +347,7 @@ def __init__(
         self.profile_connector.on_trainer_init(profiler)
 
         # init logger flags
-        self.logger_connector.on_trainer_init(logger, log_save_interval, row_log_interval)
+        self.logger_connector.on_trainer_init(logger, flush_logs_every_n_steps, log_every_n_steps)
 
         # init debugging flags
         self.debugging_connector.on_init_start(
@@ -352,7 +357,7 @@ def __init__(
             limit_test_batches,
             val_check_interval,
             overfit_batches,
-            fast_dev_run
+            fast_dev_run,
         )
 
         # set precision
@@ -502,13 +507,15 @@ def train(self):
                 met_min_steps = self.global_step >= self.min_steps if self.min_steps else True
 
                 if self.should_stop:
-                    if (met_min_epochs and met_min_steps):
+                    if met_min_epochs and met_min_steps:
                         self.train_loop.on_train_end()
                         return
                     else:
-                        log.info('Trainer was signaled to stop but required minimum epochs'
-                                 f' ({self.min_epochs}) or minimum steps ({self.min_steps}) has'
-                                 ' not been met. Training will continue...')
+                        log.info(
+                            'Trainer was signaled to stop but required minimum epochs'
+                            f' ({self.min_epochs}) or minimum steps ({self.min_steps}) has'
+                            ' not been met. Training will continue...'
+                        )
 
             # hook
             self.train_loop.on_train_end()

pytorch_lightning/trainer/training_loop.py

@@ -449,7 +449,7 @@ def on_before_backward(self, batch_idx, optimizer):
 
     def _track_gradient_norm(self):
         grad_norm_dict = {}
-        if (self.trainer.global_step + 1) % self.trainer.row_log_interval == 0:
+        if (self.trainer.global_step + 1) % self.trainer.log_every_n_steps == 0:
             if float(self.trainer.track_grad_norm) > 0:
                 model = self.trainer.get_model()
                 grad_norm_dict = model.grad_norm(self.trainer.track_grad_norm)
@@ -787,7 +787,7 @@ def build_train_args(self, batch, batch_idx, opt_idx, hiddens):
     def save_loggers_on_train_batch_end(self):
         # when loggers should save to disk
         should_save_log = (
-            (self.trainer.global_step + 1) % self.trainer.log_save_interval == 0 or self.trainer.should_stop
+            (self.trainer.global_step + 1) % self.trainer.flush_logs_every_n_steps == 0 or self.trainer.should_stop
         )
         if should_save_log or self.trainer.fast_dev_run:
             if self.trainer.is_global_zero and self.trainer.logger is not None:

tests/models/test_grad_norm.py

@@ -59,7 +59,7 @@ def test_grad_tracking(tmpdir, norm_type, rtol=5e-3):
         default_root_dir=tmpdir,
         max_epochs=3,
         track_grad_norm=norm_type,
-        row_log_interval=1,  # request grad_norms every batch
+        log_every_n_steps=1,  # request grad_norms every batch
     )
     result = trainer.fit(model)
 
@@ -76,20 +76,20 @@ def test_grad_tracking(tmpdir, norm_type, rtol=5e-3):
         assert np.allclose(log, mod, rtol=rtol)
 
 
-@pytest.mark.parametrize("row_log_interval", [1, 2, 3])
-def test_grad_tracking_interval(tmpdir, row_log_interval):
+@pytest.mark.parametrize("log_every_n_steps", [1, 2, 3])
+def test_grad_tracking_interval(tmpdir, log_every_n_steps):
     """ Test that gradient norms get tracked in the right interval and that everytime the same keys get logged. """
     trainer = Trainer(
         default_root_dir=tmpdir,
         track_grad_norm=2,
-        row_log_interval=row_log_interval,
+        log_every_n_steps=log_every_n_steps,
         max_steps=10,
     )
 
     with patch.object(trainer.logger, "log_metrics") as mocked:
         model = EvalModelTemplate()
         trainer.fit(model)
-        expected = trainer.global_step // row_log_interval
+        expected = trainer.global_step // log_every_n_steps
         grad_norm_dicts = []
         for _, kwargs in mocked.call_args_list:
             metrics = kwargs.get("metrics", {})

tests/models/test_tpu.py

@@ -266,7 +266,7 @@ def test_result_obj_on_tpu(tmpdir):
         default_root_dir=tmpdir,
         max_epochs=epochs,
         early_stop_callback=True,
-        row_log_interval=2,
+        log_every_n_steps=2,
         limit_train_batches=batches,
         weights_summary=None,
         tpu_cores=8

tests/trainer/test_correct_freq_accumulation.py

@@ -28,7 +28,7 @@ def test_training_step_scalar(tmpdir):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)

tests/trainer/test_eval_loop_flow_1_0.py

@@ -41,7 +41,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)
@@ -90,7 +90,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)
@@ -147,7 +147,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
 
@@ -211,7 +211,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
 

tests/trainer/test_eval_loop_logging_1_0.py

@@ -39,7 +39,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)
@@ -108,7 +108,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)

tests/trainer/test_train_loop_flow_dict_1_0.py

@@ -31,7 +31,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)
@@ -73,7 +73,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)
@@ -121,7 +121,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)
@@ -175,7 +175,7 @@ def backward(self, trainer, loss, optimizer, optimizer_idx):
         limit_train_batches=2,
         limit_val_batches=2,
         max_epochs=2,
-        row_log_interval=1,
+        log_every_n_steps=1,
         weights_summary=None,
     )
     trainer.fit(model)