✍️ fix some minor bugs

Author: nglehuy

setup.py

@@ -36,7 +36,7 @@
 
 setuptools.setup(
     name="TensorFlowASR",
-    version="0.7.1",
+    version="0.7.2",
     author="Huy Le Nguyen",
     author_email="[email protected]",
     description="Almost State-of-the-art Automatic Speech Recognition using Tensorflow 2",

tensorflow_asr/datasets/asr_dataset.py

@@ -140,7 +140,7 @@ def process(self, dataset: tf.data.Dataset, batch_size: int):
 
         # PREFETCH to improve speed of input length
         dataset = dataset.prefetch(AUTOTUNE)
-        self.total_steps = get_num_batches(self.total_steps, batch_size)
+        self.total_steps = get_num_batches(self.total_steps, batch_size, drop_remainders=self.drop_remainder)
         return dataset
 
     @tf.function

tensorflow_asr/datasets/keras/asr_dataset.py

@@ -93,7 +93,7 @@ def process(self, dataset, batch_size):
 
         # PREFETCH to improve speed of input length
         dataset = dataset.prefetch(AUTOTUNE)
-        self.total_steps = get_num_batches(self.total_steps, batch_size)
+        self.total_steps = get_num_batches(self.total_steps, batch_size, drop_remainders=self.drop_remainder)
         return dataset
 
 

tensorflow_asr/runners/base_runners.py

@@ -370,13 +370,12 @@ def __init__(self,
             "greed_cer": ErrorRate(func=cer, name="test_greed_cer", dtype=tf.float32)
         }
 
-    def set_output_file(self):
+    def set_output_file(self, batch_size: int = 1):
+        if not batch_size: batch_size = self.config.batch_size
         if os.path.exists(self.output_file_path):
             with open(self.output_file_path, "r", encoding="utf-8") as out:
-                self.processed_records = get_num_batches(
-                    len(out.read().splitlines()) - 1,
-                    batch_size=1
-                )
+                self.processed_records = get_num_batches(len(out.read().splitlines()) - 1, batch_size=batch_size,
+                                                         drop_remainders=False)
         else:
             with open(self.output_file_path, "w") as out:
                 out.write("PATH\tGROUNDTRUTH\tGREEDY\tBEAMSEARCH\tBEAMSEARCHLM\n")
@@ -396,7 +395,7 @@ def compile(self, trained_model: tf.keras.Model):
         self.model = trained_model
 
     def run(self, test_dataset, batch_size=None):
-        self.set_output_file()
+        self.set_output_file(batch_size=batch_size)
         self.set_test_data_loader(test_dataset, batch_size=batch_size)
         self._test_epoch()
         self._finish()