From 012d0e2b57fa10efa9a92f0f70f6385284f3ae81 Mon Sep 17 00:00:00 2001
From: antoni <antoni.s.jankowski@gmail.com>
Date: Sun, 18 Feb 2024 22:53:54 +0100
Subject: [PATCH] MIDI60: new config

related to: https://github.com/anthonio9/penn/issues/9
---
 config/ppn-split-midi60.py   | 45 ++++++++++++++++++++++++++++++++++++
 penn/data/preprocess/core.py |  7 +-----
 penn/train/core.py           |  4 ++--
 3 files changed, 48 insertions(+), 8 deletions(-)
 create mode 100644 config/ppn-split-midi60.py

diff --git a/config/ppn-split-midi60.py b/config/ppn-split-midi60.py
new file mode 100644
index 0000000..29924e9
--- /dev/null
+++ b/config/ppn-split-midi60.py
@@ -0,0 +1,45 @@
+MODULE = 'penn'
+
+# Configuration name
+CONFIG = 'ppn-split-midi60'
+
+# gset only
+DATASETS = ['gset']
+
+EVALUATION_DATASETS = ['gset']
+
+STEPS = 50000
+
+LOG_INTERVAL = 500
+
+CHECKPOINT_INTERVAL = 5000  # steps
+
+# audio parameters
+SAMPLE_RATE = 11025
+
+# the original hopsize is 256 samples, this is 4 times less than that
+HOPSIZE = 64 
+
+# use only the voiced frames
+VOICED_ONLY = True
+
+STRING_INDEX = None
+
+# poly pitch net model
+MODEL = 'ppnmidi60'
+
+PITCH_BINS = 60
+
+PITCH_CATS = 6
+
+GSET_SPLIT_PLAYERS = True
+
+REMOVE_OVERHANGS = True
+
+MIDI60 = True
+
+MIDI_OFFSET_RAND = 36
+
+GAUSSIAN_BLUR = False
+
+INTERPOLATE_UNVOICED = False
diff --git a/penn/data/preprocess/core.py b/penn/data/preprocess/core.py
index 0edd44d..9ec531d 100644
--- a/penn/data/preprocess/core.py
+++ b/penn/data/preprocess/core.py
@@ -266,8 +266,6 @@ def gset():
         pitch, times = extract_pitch_array_jams(
                 jams_track, audio_file.stem, uniform=True)
 
-        breakpoint()
-
         if penn.STRING_INDEX is not None:
             pitch = pitch[penn.STRING_INDEX, :]
             pitch = pitch[None, :]
@@ -275,7 +273,6 @@ def gset():
         if penn.INTERPOLATE_UNVOICED:
             # Fill unvoiced regions via linear interpolation
             pitch, voiced = interpolate_unvoiced(pitch)
-            breakpoint()
         else:
             unvoiced = pitch == 0
             voiced = ~unvoiced
@@ -812,9 +809,7 @@ def notes_dict_to_pitch_array60(
 
     below36 = midi_array[midi_array < 36]
 
-    if not (below36 == 0).all():
-        breakpoint()
-        pass
+    assert not (below36 == 0).all():
 
     offset_array = np.vstack(offset_list)
 
diff --git a/penn/train/core.py b/penn/train/core.py
index 742c250..67ca614 100644
--- a/penn/train/core.py
+++ b/penn/train/core.py
@@ -167,11 +167,11 @@ def train(datasets, directory, gpu=None, use_wand=False):
                         loader=test_loader,
                         gpu=gpu)
 
-                if use_wand:
+                if use_wand and not penn.MIDI60:
                     log_wandb.log({"test_logits": wandb.Image(fig)})
 
             # Evaluate
-            if step % penn.LOG_INTERVAL == 0:
+            if step % penn.LOG_INTERVAL == 0 and not penn.MIDI60:
                 evaluate_fn = functools.partial(
                     evaluate,
                     directory,