From 3562a199b371fcc91e440d2bf268b834c70c4ef2 Mon Sep 17 00:00:00 2001
From: lukewys <wuyusongwys@gmail.com>
Date: Wed, 10 Nov 2021 22:07:45 -0500
Subject: [PATCH 1/3] Update the file pattern in Urmp base dataloader to match
 the file pattern in
 https://console.cloud.google.com/storage/browser/magentadata/datasets/urmp/urmp_20210324.

---
 ddsp/training/data.py | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/ddsp/training/data.py b/ddsp/training/data.py
index fb0c4d37..74e50f68 100644
--- a/ddsp/training/data.py
+++ b/ddsp/training/data.py
@@ -411,7 +411,11 @@ def features_dict(self):
 class Urmp(TFRecordProvider):
   """Urmp training set."""
 
-  def __init__(self, base_dir, instrument_key='tpt', split='train'):
+  def __init__(self,
+               base_dir,
+               instrument_key='tpt',
+               split='train',
+               batched='batched'):
     """URMP dataset for either a specific instrument or all instruments.
 
     Args:
@@ -420,19 +424,22 @@ def __init__(self, base_dir, instrument_key='tpt', split='train'):
         ['all', 'bn', 'cl', 'db', 'fl', 'hn', 'ob', 'sax', 'tba', 'tbn',
         'tpt', 'va', 'vc', 'vn'].
       split: Choices include ['train', 'test'].
+      batched: Choices include ['batched', 'unbatched'].
     """
     self.instrument_key = instrument_key
     self.split = split
     self.base_dir = base_dir
+    self.batched = batched
     super().__init__()
 
   @property
   def default_file_pattern(self):
     if self.instrument_key == 'all':
-      file_pattern = 'all_instruments_{}.tfrecord*'.format(self.split)
+      file_pattern = 'all_instruments_{}_{}.tfrecord*'.format(
+        self.split, self.batched)
     else:
-      file_pattern = 'urmp_{}_solo_ddsp_conditioning_{}.tfrecord*'.format(
-          self.instrument_key, self.split)
+      file_pattern = 'urmp_{}_solo_ddsp_conditioning_{}_{}.tfrecord*'.format(
+        self.instrument_key, self.split, self.batched)
 
     return os.path.join(self.base_dir, file_pattern)
 

From 70def03a8a2ef1279451f2e6d327d8ca78149bcb Mon Sep 17 00:00:00 2001
From: lukewys <wuyusongwys@gmail.com>
Date: Mon, 15 Nov 2021 23:37:11 -0500
Subject: [PATCH 2/3] Change the "batched" kwarg to "suffix" which have default
 behavior same as the previous version. Add detailed documentation for the
 "suffix" kwarg.

---
 ddsp/training/data.py | 22 +++++++++++++++-------
 1 file changed, 15 insertions(+), 7 deletions(-)

diff --git a/ddsp/training/data.py b/ddsp/training/data.py
index 74e50f68..699fcf6d 100644
--- a/ddsp/training/data.py
+++ b/ddsp/training/data.py
@@ -415,7 +415,7 @@ def __init__(self,
                base_dir,
                instrument_key='tpt',
                split='train',
-               batched='batched'):
+               suffix=None):
     """URMP dataset for either a specific instrument or all instruments.
 
     Args:
@@ -424,22 +424,30 @@ def __init__(self,
         ['all', 'bn', 'cl', 'db', 'fl', 'hn', 'ob', 'sax', 'tba', 'tbn',
         'tpt', 'va', 'vc', 'vn'].
       split: Choices include ['train', 'test'].
-      batched: Choices include ['batched', 'unbatched'].
+      suffix: Choices include [None, 'batched', 'unbatched'], but broadly
+        applies to any suffix adding to the file pattern.
+        When suffix is not None, will add "_suffix" to the file pattern.
+        This option is used in gs://magentadata/datasets/urmp/urmp_20210324.
+        With the "batched" suffix, the dataloader will load tfrecords
+        containing segmented audio samples in 4 seconds. With the "unbatched"
+        suffix, the dataloader will load tfrecords containing unsegmented
+        samples which could be used for learning note sequence in URMP dataset.
+
     """
     self.instrument_key = instrument_key
     self.split = split
     self.base_dir = base_dir
-    self.batched = batched
+    self.suffix = suffix if suffix is None else '_' + suffix
     super().__init__()
 
   @property
   def default_file_pattern(self):
     if self.instrument_key == 'all':
-      file_pattern = 'all_instruments_{}_{}.tfrecord*'.format(
-        self.split, self.batched)
+      file_pattern = 'all_instruments_{}{}.tfrecord*'.format(
+        self.split, self.suffix)
     else:
-      file_pattern = 'urmp_{}_solo_ddsp_conditioning_{}_{}.tfrecord*'.format(
-        self.instrument_key, self.split, self.batched)
+      file_pattern = 'urmp_{}_solo_ddsp_conditioning_{}{}.tfrecord*'.format(
+        self.instrument_key, self.split, self.suffix)
 
     return os.path.join(self.base_dir, file_pattern)
 

From bb6aed536608396f786106758dc9ef6cb04bce70 Mon Sep 17 00:00:00 2001
From: lukewys <wuyusongwys@gmail.com>
Date: Thu, 18 Nov 2021 15:03:43 -0500
Subject: [PATCH 3/3] Fix the silly bug of f'{None}' prints 'None'. Huge
 thanks, Ethan!

---
 ddsp/training/data.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/ddsp/training/data.py b/ddsp/training/data.py
index 699fcf6d..dec02820 100644
--- a/ddsp/training/data.py
+++ b/ddsp/training/data.py
@@ -426,7 +426,7 @@ def __init__(self,
       split: Choices include ['train', 'test'].
       suffix: Choices include [None, 'batched', 'unbatched'], but broadly
         applies to any suffix adding to the file pattern.
-        When suffix is not None, will add "_suffix" to the file pattern.
+        When suffix is not None, will add "_{suffix}" to the file pattern.
         This option is used in gs://magentadata/datasets/urmp/urmp_20210324.
         With the "batched" suffix, the dataloader will load tfrecords
         containing segmented audio samples in 4 seconds. With the "unbatched"
@@ -437,7 +437,7 @@ def __init__(self,
     self.instrument_key = instrument_key
     self.split = split
     self.base_dir = base_dir
-    self.suffix = suffix if suffix is None else '_' + suffix
+    self.suffix = '' if suffix is None else '_' + suffix
     super().__init__()
 
   @property