Change all prints and nemo.logging to from nemo import logging; loggi…

…ng (#311) * change print and nemo.logging to from nemo import logging; logging. Signed-off-by: Jason <[email protected]> * style fix Signed-off-by: Jason <[email protected]> * change some infos to warnings Signed-off-by: Jason <[email protected]> * fix lgtm suggestions Signed-off-by: Jason <[email protected]> * abs path Signed-off-by: Jason <[email protected]>
NVIDIA · Jan 31, 2020 · 4137c2b · 4137c2b
1 parent 1fddca2
commit 4137c2b
Show file tree

Hide file tree

Showing 54 changed files with 341 additions and 279 deletions.
diff --git a/docs/docs_zh/sources/source/nlp/ner.rst b/docs/docs_zh/sources/source/nlp/ner.rst
@@ -149,7 +149,7 @@ text.txt 每一行包含文本序列，其中词以空格来进行分隔。label
 
         callback_train = nemo.core.SimpleLossLoggerCallback(
             tensors=[loss],
-            print_func=lambda x: print("Loss: {:.3f}".format(x[0].item())))
+            print_func=lambda x: logging.info("Loss: {:.3f}".format(x[0].item())))
 
         train_data_size = len(train_data_layer)
 

diff --git a/docs/docs_zh/sources/source/tutorials/callbacks.rst b/docs/docs_zh/sources/source/tutorials/callbacks.rst
@@ -24,13 +24,14 @@ SimpleLossLoggerCallback 是用来记录训练过程中的一些指标数据比
 get_tb_values() 和 log_to_tb_func() 函数的输入。两个推荐重写的参数是 print_func() 和
 get_tb_values() 或者 log_to_tb_func() 任选其一。
 
-print_func() 应该用来记录打印到屏幕上的值。我们推荐使用 nemo.logging.info()
+print_func() 应该用来记录打印到屏幕上的值。我们推荐使用 logging.info()
 来取代 print() 函数。比如，可以这么打印 loss 值：
 
 .. code-block:: python
 
+    from nemo import logging
     def my_print_func(tensors):
-        nemo.logging.info(f"Loss {tensors[0]}")
+        logging.info(f"Loss {tensors[0]}")
 
 我们提供了两个方法来打印到 tensorboard: get_tb_values() 和
 log_to_tb_func()。对于记录标量的简单用例，我们推荐使用 get_tb_values()。

diff --git a/docs/docs_zh/sources/source/tutorials/complex_training.rst b/docs/docs_zh/sources/source/tutorials/complex_training.rst
@@ -42,7 +42,7 @@ NeMo 进一步扩充了用例，这些用例会用到多个损失函数和多个
     # 更新打印函数加入两个损失函数的张量
     callback = nemo.core.SimpleLossLoggerCallback(
         tensors=[l1_loss_tensor, mse_loss_tensor],
-        print_func=lambda x: print(
+        print_func=lambda x: logging.info(
             f'Train Loss: {str(x[0].item() + x[1].item())}')
     )
 
@@ -80,7 +80,7 @@ NeMo 进一步扩充了用例，这些用例会用到多个损失函数和多个
     # SimpleLossLoggerCallback 把损失函数值打印到控制台
     callback = nemo.core.SimpleLossLoggerCallback(
         tensors=[l1_loss_tensor, mse_loss_tensor],
-        print_func=lambda x: print(
+        print_func=lambda x: logging.info(
             f'L1 Loss: {str(x[0].item())}'
             f'MSE Loss: {str(x[1].item())}')
     )

diff --git a/docs/docs_zh/sources/source/tutorials/examples.rst b/docs/docs_zh/sources/source/tutorials/examples.rst
@@ -35,7 +35,7 @@ Hello World
     # SimpleLossLoggerCallback 打印损失函数值到控制台
     callback = nemo.core.SimpleLossLoggerCallback(
         tensors=[lss],
-        print_func=lambda x: print(f'Train Loss: {str(x[0].item())}'))
+        print_func=lambda x: logging.info(f'Train Loss: {str(x[0].item())}'))
 
     # 触发“训练”操作
     nf.train([lss], callbacks=[callback],
@@ -129,9 +129,9 @@ Hello World
         source = ' '.join([s for s in source if s != 'EOS' and s != 'PAD'])
         response = ' '.join([s for s in response if s != 'EOS' and s != 'PAD'])
         target = ' '.join([s for s in target if s != 'EOS' and s != 'PAD'])
-        print(f"Train Loss:{str(tensors[0].item())}")
-        print(f"SOURCE: {source} <---> PREDICTED RESPONSE: {response} "
-              f"<---> TARGET: {target}")
+        logging.info(f"Train Loss:{str(tensors[0].item())}")
+        logging.info(f"SOURCE: {source} <---> PREDICTED RESPONSE: {response} "
+                     f"<---> TARGET: {target}")
 
 
     callback = nemo.core.SimpleLossLoggerCallback(

diff --git a/docs/sources/source/nlp/joint_intent_slot_filling.rst b/docs/sources/source/nlp/joint_intent_slot_filling.rst
@@ -109,7 +109,7 @@ Next, we define all Neural Modules participating in our joint intent slot fillin
                             num_gpus=1,
                             local_rank=0,
                             mode='train'):
-            nemo.logging.info(f"Loading {mode} data...")
+            logging.info(f"Loading {mode} data...")
             data_file = f'{data_desc.data_dir}/{mode}.tsv'
             slot_file = f'{data_desc.data_dir}/{mode}_slots.tsv'
             shuffle = args.shuffle_data if mode == 'train' else False
@@ -133,15 +133,15 @@ Next, we define all Neural Modules participating in our joint intent slot fillin
                 subtokens_mask, intents, slots = data_layer()
             data_size = len(data_layer)
 
-            print(f'The length of data layer is {data_size}')
+            logging.info(f'The length of data layer is {data_size}')
 
             if data_size < batch_size:
-                nemo.logging.warning("Batch_size is larger than the dataset size")
-                nemo.logging.warning("Reducing batch_size to dataset size")
+                logging.warning("Batch_size is larger than the dataset size")
+                logging.warning("Reducing batch_size to dataset size")
                 batch_size = data_size
 
             steps_per_epoch = math.ceil(data_size / (batch_size * num_gpus))
-            nemo.logging.info(f"Steps_per_epoch = {steps_per_epoch}")
+            logging.info(f"Steps_per_epoch = {steps_per_epoch}")
 
             hidden_states = pretrained_bert_model(input_ids=ids,
                                                   token_type_ids=type_ids,

diff --git a/docs/sources/source/nlp/ner.rst b/docs/sources/source/nlp/ner.rst
@@ -137,7 +137,7 @@ Now, we will set up our callbacks. We will use 3 callbacks:
 
         callback_train = nemo.core.SimpleLossLoggerCallback(
             tensors=[loss],
-            print_func=lambda x: print("Loss: {:.3f}".format(x[0].item())))
+            print_func=lambda x: logging.info("Loss: {:.3f}".format(x[0].item())))
 
         train_data_size = len(train_data_layer)
 

diff --git a/docs/sources/source/nlp/punctuation.rst b/docs/sources/source/nlp/punctuation.rst
@@ -204,7 +204,7 @@ Now, we will set up our callbacks. We will use 3 callbacks:
 
         callback_train = nemo.core.SimpleLossLoggerCallback(
         tensors=[task_loss, punct_loss, capit_loss, punct_logits, capit_logits],
-        print_func=lambda x: print("Loss: {:.3f}".format(x[0].item())),
+        print_func=lambda x: logging.info("Loss: {:.3f}".format(x[0].item())),
         step_freq=STEP_FREQ)
 
         train_data_size = len(train_data_layer)
@@ -293,7 +293,7 @@ Run inference, append punctuation and capitalize words based on the generated pr
     capit_preds = np.argmax(capit_logits, axis=2)
 
     for i, query in enumerate(queries):
-        nemo.logging.info(f'Query: {query}')
+        logging.info(f'Query: {query}')
 
         punct_pred = punct_preds[i][subtokens_mask[i] > 0.5]
         capit_pred = capit_preds[i][subtokens_mask[i] > 0.5]
@@ -312,7 +312,7 @@ Run inference, append punctuation and capitalize words based on the generated pr
             if punct_label != 'O':
                 output += punct_label
             output += ' '
-        nemo.logging.info(f'Combined: {output.strip()}\n')
+        logging.info(f'Combined: {output.strip()}\n')
 
 Inference results:
 

diff --git a/docs/sources/source/nlp/question_answering.rst b/docs/sources/source/nlp/question_answering.rst
@@ -157,7 +157,7 @@ Next, we define all Neural Modules participating in our question answering class
 
         train_callback = nemo.core.SimpleLossLoggerCallback(
             tensors=train_tensors,
-            print_func=lambda x: print("Loss: {:.3f}".format(x[0].item())),
+            print_func=lambda x: logging.info("Loss: {:.3f}".format(x[0].item())),
             get_tb_values=lambda x: [["loss", x[0]]],
             step_freq=args.step_freq,
             tb_writer=neural_factory.tb_writer)

diff --git a/docs/sources/source/tutorials/callbacks.rst b/docs/sources/source/tutorials/callbacks.rst
@@ -28,13 +28,14 @@ training. The two reccomended arguments to override are print_func(), and
 either get_tb_values() or log_to_tb_func().
 
 print_func() should be used to log values to screen. We recommend using
-nemo.logging.info() in place
+logging.info() in place
 of print(). For example, it can be used to print the loss value:
 
 .. code-block:: python
 
+    from nemo import logging
     def my_print_func(tensors):
-        nemo.logging.info(f"Loss {tensors[0]}")
+        logging.info(f"Loss {tensors[0]}")
 
 We provide two methods to log to tensorboard: get_tb_values() and
 log_to_tb_func(). For simple use case of logging scalars, we recommend

diff --git a/docs/sources/source/tutorials/complex_training.rst b/docs/sources/source/tutorials/complex_training.rst
@@ -44,7 +44,7 @@ Here's an example:
     # Update printing function to add both losses
     callback = nemo.core.SimpleLossLoggerCallback(
         tensors=[l1_loss_tensor, mse_loss_tensor],
-        print_func=lambda x: print(
+        print_func=lambda x: logging.info(
             f'Train Loss: {str(x[0].item() + x[1].item())}')
     )
 
@@ -84,7 +84,7 @@ loop.
     # SimpleLossLoggerCallback will print loss values to console.
     callback = nemo.core.SimpleLossLoggerCallback(
         tensors=[l1_loss_tensor, mse_loss_tensor],
-        print_func=lambda x: print(
+        print_func=lambda x: logging.info(
             f'L1 Loss: {str(x[0].item())}'
             f'MSE Loss: {str(x[1].item())}')
     )

diff --git a/docs/sources/source/tutorials/examples.rst b/docs/sources/source/tutorials/examples.rst
@@ -35,7 +35,7 @@ This example shows how to build a model which learn Taylor's coefficients for y=
     # SimpleLossLoggerCallback will print loss values to console.
     callback = nemo.core.SimpleLossLoggerCallback(
         tensors=[lss],
-        print_func=lambda x: print(f'Train Loss: {str(x[0].item())}'))
+        print_func=lambda x: logging.info(f'Train Loss: {str(x[0].item())}'))
 
     # Invoke "train" action
     nf.train([lss], callbacks=[callback],
@@ -129,9 +129,9 @@ During training model will print:
         source = ' '.join([s for s in source if s != 'EOS' and s != 'PAD'])
         response = ' '.join([s for s in response if s != 'EOS' and s != 'PAD'])
         target = ' '.join([s for s in target if s != 'EOS' and s != 'PAD'])
-        print(f"Train Loss:{str(tensors[0].item())}")
-        print(f"SOURCE: {source} <---> PREDICTED RESPONSE: {response} "
-              f"<---> TARGET: {target}")
+        logging.info(f"Train Loss:{str(tensors[0].item())}")
+        logging.info(f"SOURCE: {source} <---> PREDICTED RESPONSE: {response} "
+                     f"<---> TARGET: {target}")
 
 
     callback = nemo.core.SimpleLossLoggerCallback(

diff --git a/examples/applications/asr_service/app/__init__.py b/examples/applications/asr_service/app/__init__.py
@@ -8,6 +8,8 @@
 import nemo
 import nemo.collections.asr as nemo_asr
 
+logging = nemo.logging
+
 app = Flask(__name__)
 # make sure WORK_DIR exists before calling your service
 # in this folder, the service will store received .wav files and constructed
@@ -46,7 +48,7 @@
         vocab=labels, beam_width=64, alpha=2.0, beta=1.0, lm_path=LM_PATH, num_cpus=max(os.cpu_count(), 1),
     )
 else:
-    print("Beam search is not enabled")
+    logging.info("Beam search is not enabled")
 
 if __name__ == '__main__':
     app.run()
diff --git a/examples/applications/asr_service/app/routes.py b/examples/applications/asr_service/app/routes.py
@@ -20,10 +20,12 @@
 import nemo
 import nemo.collections.asr as nemo_asr
 
+logging = nemo.logging
+
 try:
     from app import beam_search_with_lm
 except ImportError:
-    print("Not using Beam Search Decoder with LM")
+    logging.info("Not using Beam Search Decoder with LM")
     ENABLE_NGRAM = False
 
 
@@ -46,7 +48,7 @@ def wav_to_text(manifest, greedy=True):
     predictions = greedy_decoder(log_probs=log_probs)
 
     if ENABLE_NGRAM:
-        print('Running with beam search')
+        logging.info('Running with beam search')
         beam_predictions = beam_search_with_lm(log_probs=log_probs, log_probs_length=encoded_len)
         eval_tensors = [beam_predictions]
 

diff --git a/examples/asr/jasper_an4.py b/examples/asr/jasper_an4.py
@@ -20,6 +20,8 @@
 )
 from nemo.utils.lr_policies import CosineAnnealing
 
+logging = nemo.logging
+
 
 def create_dags(jasper_params, args, nf):
     vocab = jasper_params['labels']
@@ -37,7 +39,7 @@ def create_dags(jasper_params, args, nf):
     num_samples = len(data_layer)
     steps_per_epoch = math.ceil(num_samples / (args.batch_size * args.iter_per_step * nf.world_size))
     total_steps = steps_per_epoch * args.num_epochs
-    print("Train samples=", num_samples, "num_steps=", total_steps)
+    logging.info("Train samples=", num_samples, "num_steps=", total_steps)
 
     data_preprocessor = nemo_asr.AudioToMelSpectrogramPreprocessor(
         **jasper_params["AudioToMelSpectrogramPreprocessor"]
@@ -57,7 +59,7 @@ def create_dags(jasper_params, args, nf):
     )
 
     num_samples = len(data_layer_eval)
-    nemo.logging.info(f"Eval samples={num_samples}")
+    logging.info(f"Eval samples={num_samples}")
 
     jasper_encoder = nemo_asr.JasperEncoder(**jasper_params["JasperEncoder"])
 
@@ -84,7 +86,7 @@ def create_dags(jasper_params, args, nf):
     loss_e = ctc_loss(
         log_probs=log_probs_e, targets=transcript_e, input_length=encoded_len_e, target_length=transcript_len_e,
     )
-    nemo.logging.info("Num of params in encoder: {0}".format(jasper_encoder.num_weights))
+    logging.info("Num of params in encoder: {0}".format(jasper_encoder.num_weights))
 
     # Callbacks to print info to console and Tensorboard
     train_callback = nemo.core.SimpleLossLoggerCallback(
@@ -194,10 +196,10 @@ def main():
     )
 
     if args.test_after_training:
-        nemo.logging.info("Testing greedy and beam search with LM WER.")
+        logging.info("Testing greedy and beam search with LM WER.")
         # Create BeamSearch NM
         if nf.world_size > 1:
-            nemo.logging.warning("Skipping beam search WER as it does not " "work if doing distributed training.")
+            logging.warning("Skipping beam search WER as it does not " "work if doing distributed training.")
         else:
             beam_search_with_lm = nemo_asr.BeamSearchDecoderWithLM(
                 vocab=vocab, beam_width=64, alpha=2.0, beta=1.5, lm_path=args.lm, num_cpus=max(os.cpu_count(), 1),
@@ -210,7 +212,7 @@ def main():
             greedy_hypotheses = post_process_predictions(evaluated_tensors[1], vocab)
             references = post_process_transcripts(evaluated_tensors[2], evaluated_tensors[3], vocab)
             wer = word_error_rate(hypotheses=greedy_hypotheses, references=references)
-            nemo.logging.info("Greedy WER: {:.2f}%".format(wer * 100))
+            logging.info("Greedy WER: {:.2f}%".format(wer * 100))
             if wer > wer_thr:
                 nf.sync_all_processes(False)
                 raise ValueError(f"Final eval greedy WER {wer * 100:.2f}% > :" f"than {wer_thr * 100:.2f}%")
@@ -225,7 +227,7 @@ def main():
                     beam_hypotheses.append(j[0][1])
 
             beam_wer = word_error_rate(hypotheses=beam_hypotheses, references=references)
-            nemo.logging.info("Beam WER {:.2f}%".format(beam_wer * 100))
+            logging.info("Beam WER {:.2f}%".format(beam_wer * 100))
             assert beam_wer <= beam_wer_thr, "Final eval beam WER {:.2f}%  > than {:.2f}%".format(
                 beam_wer * 100, beam_wer_thr * 100
             )
@@ -266,7 +268,7 @@ def main():
             greedy_hypotheses = post_process_predictions(evaluated_tensors[1], vocab)
             references = post_process_transcripts(evaluated_tensors[2], evaluated_tensors[3], vocab)
             wer_new = word_error_rate(hypotheses=greedy_hypotheses, references=references)
-            nemo.logging.info("New greedy WER: {:.2f}%".format(wer_new * 100))
+            logging.info("New greedy WER: {:.2f}%".format(wer_new * 100))
             if wer_new > wer * 1.1:
                 nf.sync_all_processes(False)
                 raise ValueError(

diff --git a/examples/image/transfer_learning.py b/examples/image/transfer_learning.py
@@ -9,16 +9,18 @@
 import nemo
 from nemo.backends.pytorch.torchvision.helpers import compute_accuracy, eval_epochs_done_callback, eval_iter_callback
 
+logging = nemo.logging
+
 sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '../..')))
 
 if not os.path.isdir("hymenoptera_data"):
-    print("Datafolder not found. Downloading data from the Web")
+    logging.info("Datafolder not found. Downloading data from the Web")
     subprocess.run(["wget", "https://download.pytorch.org/tutorial/hymenoptera_data.zip"])
     zip_ref = zipfile.ZipFile('hymenoptera_data.zip', 'r')
     zip_ref.extractall('.')
     zip_ref.close()
 else:
-    print("Found data folder - hymenoptera_data")
+    logging.info("Found data folder - hymenoptera_data")
 
 parser = argparse.ArgumentParser(description='Transfer Learning')
 parser.add_argument("--batch_size", default=32, type=int)

diff --git a/examples/start_here/chatbot_example.py b/examples/start_here/chatbot_example.py
@@ -4,6 +4,8 @@
 
 import nemo
 
+logging = nemo.logging
+
 # Get Data
 data_file = "movie_data.txt"
 if not os.path.isfile(data_file):
@@ -67,8 +69,8 @@ def outputs2words(tensors, vocab):
     source = ' '.join([s for s in source if s != 'EOS' and s != 'PAD'])
     response = ' '.join([s for s in response if s != 'EOS' and s != 'PAD'])
     target = ' '.join([s for s in target if s != 'EOS' and s != 'PAD'])
-    print(f"Train Loss:{str(tensors[0].item())}")
-    print(f"SOURCE: {source} <---> PREDICTED RESPONSE: {response} " f"<---> TARGET: {target}")
+    logging.info(f"Train Loss:{str(tensors[0].item())}")
+    logging.info(f"SOURCE: {source} <---> PREDICTED RESPONSE: {response} " f"<---> TARGET: {target}")
 
 
 callback = nemo.core.SimpleLossLoggerCallback(

diff --git a/examples/start_here/chatbot_example2.py b/examples/start_here/chatbot_example2.py
@@ -5,6 +5,8 @@
 
 import nemo
 
+logging = nemo.logging
+
 # Get Data
 data_file = "movie_data.txt"
 if not os.path.isfile(data_file):
@@ -74,7 +76,7 @@ def outputs2words(tensors, vocab):
     source = ' '.join([s for s in source if s != 'EOS' and s != 'PAD'])
     response = ' '.join([s for s in response if s != 'EOS' and s != 'PAD'])
     target = ' '.join([s for s in target if s != 'EOS' and s != 'PAD'])
-    print(f'Train Loss: {str(tensors[0].item())}')
+    logging.info(f'Train Loss: {str(tensors[0].item())}')
     tmp = " SOURCE: {0} <---> PREDICTED RESPONSE: {1} <---> TARGET: {2}"
     return tmp.format(source, response, target)
 

diff --git a/examples/start_here/simplest_example.py b/examples/start_here/simplest_example.py
@@ -1,6 +1,8 @@
 # Copyright (c) 2019 NVIDIA Corporation
 import nemo
 
+logging = nemo.logging
+
 nf = nemo.core.NeuralModuleFactory()
 # To use CPU-only do:
 # from nemo.core import DeviceType
@@ -19,7 +21,7 @@
 
 # SimpleLossLoggerCallback will print loss values to console.
 callback = nemo.core.SimpleLossLoggerCallback(
-    tensors=[lss], print_func=lambda x: print(f'Train Loss: {str(x[0].item())}'),
+    tensors=[lss], print_func=lambda x: logging.info(f'Train Loss: {str(x[0].item())}'),
 )
 
 # Invoke "train" action