create_duration_tuples can create overlapping tuples

.github/workflows/pytest.yml

@@ -15,9 +15,9 @@ jobs:
       matrix:
         python-version: ["3.8", "3.9", "3.10", "3.11"]  # tensorflow not supported with 3.12
         os: [ ubuntu-latest, windows-latest ]
-        #include:  # TODO uncomment when multithreading pytest-xdist is fixed (PR #142)
-        #  - python-version: "3.11"
-        #    os: macos-latest
+        include:
+          - python-version: "3.11"
+            os: macos-latest
 
     steps:
       - uses: actions/checkout@v4

docs/conf.py

@@ -16,7 +16,7 @@
 project = "MidiTok"
 copyright = "2023, Nathan Fradet"  # noqa: A001
 author = "Nathan Fradet"
-release = "3.0.1"
+release = "3.0.2"
 
 # -- General configuration ---------------------------------------------------
 # https://www.sphinx-doc.org/en/master/usage/configuration.html#general-configuration

miditok/midi_tokenizer.py

@@ -1952,15 +1952,30 @@ def __create_durations_tuples(self) -> list[tuple[int, int, int]]:
                 (beat, pos, beat_res)
                 for beat in range(*beat_range)
                 for pos in range(beat_res)
+                if not (beat == 0 and pos == 0)
             ]
-        durations += [
+        durations.append(
             (
                 max(max(self.config.beat_res)),
                 0,
                 self.config.beat_res[max(self.config.beat_res)],
             )
-        ]  # the last one
-        del durations[0]  # removes duration of 0
+        )  # adds the last one
+
+        # Sort them
+        durations.sort(key=lambda dt: dt[0] + dt[1] / dt[2])
+
+        # Deduplicate identical values
+        # For two identical values, we only keep the tuple having the lowest res
+        i = 0
+        while i < len(durations) - 1:
+            dur_i_b, dur_i_p, dur_i_r = durations[i]
+            dur_ip_b, dur_ip_p, dur_ip_r = durations[i + 1]
+            if (dur_i_b + dur_i_p / dur_i_r) == (dur_ip_b + dur_ip_p / dur_ip_r):
+                del durations[i if dur_i_r > dur_ip_r else i + 1]
+            else:
+                i += 1
+
         return durations
 
     def __create_tpb_per_ts(self) -> dict[int, int]:

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "miditok"
-version = "3.0.1"
+version = "3.0.2"
 description = "MIDI / symbolic music tokenizers for Deep Learning models."
 readme = "README.md"
 license = "MIT"