[DRAFT] Reorganize repository to a somewhat standardized format

.gitignore

@@ -3,11 +3,10 @@
 __pycache__
 .python-version
 
-# Datasets
+# Directories
 datasets/
-
-# Outputs
-runs
+runs/
+reference_models/heidenreich/datasets/
 
 # Miscellaneous
 **/.DS_Store

pyproject.toml

@@ -0,0 +1,52 @@
+[build-system]
+requires = ["hatchling", "hatch-requirements-txt"]
+build-backend = "hatchling.build"
+
+[project]
+name = "timbre"
+version = "0"
+# dependencies = []
+requires-python = ">=3.8"
+authors = [{ name = "Quinn Ouyang", email = "qouyang3+timbre@illinois.edu" }]
+# maintainers = [
+#     { name = "Quinn qouyang", email = "qouyang3+timbre@illinois.edu" },
+# ]
+# description = ""
+readme = "README.md"
+# license = {file = "LICENSE"}
+# keywords = []
+classifiers = [
+    "Development Status :: 1 - Planning",
+    "Intended Audience :: Education",
+    "Intended Audience :: Science/Research",
+    # "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3.8",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Multimedia :: Sound/Audio :: Sound Synthesis",
+]
+dynamic = ["dependencies"]
+
+[project.urls]
+# homepage = ""
+source = "https://github.com/quinnouyang/timbre"
+github = "https://github.com/quinnouyang/timbre.git"
+# download = ""
+# changelog = ""
+# documentation = ""
+issues = "https://github.com/quinnouyang/timbre/issues"
+
+[tool.hatch.metadata]
+allow-direct-references = true
+
+[tool.hatch.metadata.hooks.requirements_txt]
+files = ["requirements.txt"]
+
+# [tool.hatch.metadata.hooks.requirements_txt.optional-dependencies]
+# cli = ["requirements-dev.txt"]
+
+[tool.black]
+extend-exclude = "reference_models"
+
+[tool.isort]
+profile = "black"
+skip_glob = ["reference_models"]

requirements.txt

@@ -1,9 +1,12 @@
 black
 ipykernel
+isort
+librosa
 matplotlib
-scipy
+pip
 tensorboard
-torch
+torch_tb_profiler
 torchaudio
 torchvision
 tqdm
+yapecs

timbre/__init__.py

@@ -1,2 +1,18 @@
-from .model.train.utils import train, test, plot
-from .model.vae import VAE, VAEOutput
+###############################################################################
+# Configuration
+###############################################################################
+
+
+import yapecs
+
+from .config import defaults
+
+yapecs.configure("timbre", defaults)
+
+###############################################################################
+# Module
+###############################################################################
+
+# from .train import ...
+from . import data, model
+from .config.defaults import *

timbre/config/defaults.py

@@ -0,0 +1,99 @@
+# Based off of https://github.com/maxrmorrison/deep-learning-project-template/blob/main/NAME/config/defaults.py
+
+from datetime import datetime
+from pathlib import Path
+
+from torch.utils.tensorboard.writer import SummaryWriter
+
+from .utils import get_device, should_pin_memory
+
+# Configuration name
+CONFIG = "defaults"
+
+
+###############################################################################
+# Data
+###############################################################################
+
+# Dataset names
+DATASETS = ["nsynth"]
+
+# EVALUATION_DATASETS = DATASETS
+
+
+###############################################################################
+# Directories
+###############################################################################
+
+_PACKAGE_DIR = Path(__file__).parent.parent
+
+# For assets to bundle with pip release
+# ASSETS_DIR = _PACKAGE_DIR / "assets"
+
+_ROOT_DIR = _PACKAGE_DIR.parent
+
+_DATA_DIR = _ROOT_DIR / "datasets"
+
+# For preprocessed features
+# CACHE_DIR = _DATA_DIR / "cache"
+
+# For unprocessed datasets
+SOURCES_DIR = _DATA_DIR / "sources"
+
+# For preprocessed datasets
+PREPROCESSED_DIR = _DATA_DIR / "preprocessed"
+
+# For training and adaptation artifacts
+RUNS_DIR = _ROOT_DIR / "runs"
+
+# For evaluation artifacts
+# EVAL_DIR = _ROOT_DIR.parent / "eval"
+
+
+###############################################################################
+# Training
+###############################################################################
+
+
+# Batch size per gpu
+BATCH_SIZE = 128
+
+# Steps between saving checkpoints
+# CHECKPOINT_INTERVAL = 25000
+
+# Training steps
+# STEPS = 300000
+
+# Worker threads for data loading
+NUM_WORKERS = 8
+
+# RANDOM_SEED = 1234
+
+# [TODO] Formalize these options
+
+LEARN_RATE = 1e-3
+WEIGHT_DECAY = 1e-2
+NUM_EPOCHS = 64
+# INPUT_DIM = 16064
+INPUT_DIM = 2048
+LATENT_DIM = 2
+# HIDDEN_DIM = 8064
+HIDDEN_DIM = 1024
+
+DEVICE = get_device()
+
+USE_PIN_MEMORY = should_pin_memory()
+DATETIME_NOW = datetime.now().strftime("%Y%m%d-%H%M%S")
+WRITER = SummaryWriter(RUNS_DIR / f"log_{DATETIME_NOW}")
+
+
+###############################################################################
+# Evaluation
+###############################################################################
+
+
+# Steps between tensorboard logging
+# EVALUATION_INTERVAL = 2500  # steps
+
+# Steps to perform for tensorboard logging
+# DEFAULT_EVALUATION_STEPS = 16

timbre/config/malleus.py

@@ -0,0 +1,14 @@
+from pathlib import Path
+
+MODULE = "timbre"
+
+# Configuration name
+CONFIG = "malleus"
+
+_DATA_DIR = Path("/") / "mnt" / "data" / "quinn" / "timbre"
+
+# For unprocessed datasets
+SOURCES_DIR = _DATA_DIR / "sources"
+
+# For preprocessed features
+PREPROCESSED_DIR = _DATA_DIR / "preprocessed"

timbre/config/utils.py

@@ -0,0 +1,13 @@
+import torch
+
+
+def get_device() -> torch.device:
+    return torch.device(
+        "cuda"
+        if torch.cuda.is_available()
+        else "mps" if torch.backends.mps.is_available() else "cpu"
+    )
+
+
+def should_pin_memory() -> bool:
+    return get_device() == "cuda"

data/utils/nsynth.py → timbre/data/nsynth.py

@@ -1,12 +1,12 @@
 import json
-
 from argparse import ArgumentParser
 from contextlib import suppress
-from os import path, listdir
+from os import listdir, path
 from pathlib import Path
+from typing import Any, Literal, NamedTuple
+
 from torch import Tensor, float32
 from torch.utils.data import Dataset
-from typing import NamedTuple, Literal, Any
 from torchaudio import load, transforms
 
 transform_melspec = transforms.MelSpectrogram(n_fft=512, n_mels=64)
@@ -43,17 +43,17 @@ class NSynthDataset(Dataset):
 
     Parameters
     ----------
-    `annotations_file` : `str` | `Path`
-        Path to the JSON file containing the annotations
-    `audio_dir` : `str` | `Path`
-        Path to the directory containing the audio files
+    `source_dir` : `str` | `Path`
+        Path to a subset of the unprocessed NSynth dataset directory containing `examples.json` and `audio` directory
     """
 
-    def __init__(
-        self,
-        annotations_file: str | Path,
-        audio_dir: str | Path,
-    ) -> None:
+    def __init__(self, source_dir: str | Path) -> None:
+        if isinstance(source_dir, str):
+            source_dir = Path(source_dir)
+
+        annotations_file = source_dir / "examples.json"
+        audio_dir = source_dir / "audio"
+
         with open(annotations_file, "r") as f:
             self.annotations: dict[str, Any] = json.load(f)
             self.keys = sorted(self.annotations.keys())  # note_strs
@@ -109,9 +109,6 @@ def __getitem__(self, i: int) -> Tensor:
 
     print(f"TEST: Loading NSynthDataset and printing its last example")
 
-    D = NSynthDataset(
-        path.join(args.subset_path, "examples.json"),
-        path.join(args.subset_path, "audio"),
-    )
+    D = NSynthDataset(args.subset_path)
 
     print(D[-1])