Add parametric UMAP model family

pyproject.toml

@@ -31,6 +31,7 @@ dependencies = [
     "pytorch-lightning >=1.8.4.post0, <2.0",
     "matplotlib >=3.3.3",
     "numpy >=1.18.1",
+    "pynndescent >=0.5.10",
     "scipy >=1.4.1",
     "SoundFile >=0.10.3",
     "pandas >=1.0.1",
@@ -39,6 +40,7 @@ dependencies = [
     "torch >=1.7.1, <2.0.0",
     "torchvision >=0.5.0",
     "tqdm >=4.42.1",
+    "umap-learn >=0.5.3",
 ]
 
 [project.optional-dependencies]

src/vak/cli/eval.py

@@ -58,8 +58,10 @@ def eval(toml_path):
         checkpoint_path=cfg.eval.checkpoint_path,
         labelmap_path=cfg.eval.labelmap_path,
         output_dir=cfg.eval.output_dir,
-        window_size=cfg.dataloader.window_size,
         num_workers=cfg.eval.num_workers,
+        batch_size=cfg.eval.batch_size,
+        transform_params=cfg.eval.transform_params,
+        dataset_params=cfg.eval.dataset_params,
         spect_scaler_path=cfg.eval.spect_scaler_path,
         device=cfg.eval.device,
         post_tfm_kwargs=cfg.eval.post_tfm_kwargs,

src/vak/cli/learncurve.py

@@ -61,10 +61,13 @@ def learning_curve(toml_path):
         model_name=model_name,
         model_config=model_config,
         dataset_path=cfg.learncurve.dataset_path,
-        window_size=cfg.dataloader.window_size,
         batch_size=cfg.learncurve.batch_size,
         num_epochs=cfg.learncurve.num_epochs,
         num_workers=cfg.learncurve.num_workers,
+        train_transform_params=cfg.learncurve.train_transform_params,
+        train_dataset_params=cfg.learncurve.train_dataset_params,
+        val_transform_params=cfg.learncurve.val_transform_params,
+        val_dataset_params=cfg.learncurve.val_dataset_params,
         results_path=results_path,
         post_tfm_kwargs=cfg.learncurve.post_tfm_kwargs,
         normalize_spectrograms=cfg.learncurve.normalize_spectrograms,

src/vak/cli/predict.py

@@ -52,9 +52,9 @@ def predict(toml_path):
         dataset_path=cfg.predict.dataset_path,
         checkpoint_path=cfg.predict.checkpoint_path,
         labelmap_path=cfg.predict.labelmap_path,
-        window_size=cfg.dataloader.window_size,
         num_workers=cfg.predict.num_workers,
-        spect_key=cfg.spect_params.spect_key,
+        transform_params=cfg.predict.transform_params,
+        dataset_params=cfg.predict.dataset_params,
         timebins_key=cfg.spect_params.timebins_key,
         spect_scaler_path=cfg.predict.spect_scaler_path,
         device=cfg.predict.device,

src/vak/cli/train.py

@@ -61,7 +61,10 @@ def train(toml_path):
         model_name=model_name,
         model_config=model_config,
         dataset_path=cfg.train.dataset_path,
-        window_size=cfg.dataloader.window_size,
+        train_transform_params=cfg.train.train_transform_params,
+        train_dataset_params=cfg.train.train_dataset_params,
+        val_transform_params=cfg.train.val_transform_params,
+        val_dataset_params=cfg.train.val_dataset_params,
         batch_size=cfg.train.batch_size,
         num_epochs=cfg.train.num_epochs,
         num_workers=cfg.train.num_workers,

src/vak/config/__init__.py

@@ -1,7 +1,6 @@
 """sub-package that parses config.toml files and returns config object"""
 from . import (
     config,
-    dataloader,
     eval,
     learncurve,
     model,

src/vak/config/config.py

@@ -1,7 +1,6 @@
 import attr
 from attr.validators import instance_of, optional
 
-from .dataloader import DataLoaderConfig
 from .eval import EvalConfig
 from .learncurve import LearncurveConfig
 from .predict import PredictConfig
@@ -20,8 +19,6 @@ class Config:
         represents ``[PREP]`` section of config.toml file
     spect_params : vak.config.spect_params.SpectParamsConfig
         represents ``[SPECT_PARAMS]`` section of config.toml file
-    dataloader : vak.config.dataloader.DataLoaderConfig
-        represents ``[DATALOADER]`` section of config.toml file
     train : vak.config.train.TrainConfig
         represents ``[TRAIN]`` section of config.toml file
     eval : vak.config.eval.EvalConfig
@@ -31,14 +28,9 @@ class Config:
     learncurve : vak.config.learncurve.LearncurveConfig
         represents ``[LEARNCURVE]`` section of config.toml file
     """
-
     spect_params = attr.ib(
         validator=instance_of(SpectParamsConfig), default=SpectParamsConfig()
     )
-    dataloader = attr.ib(
-        validator=instance_of(DataLoaderConfig), default=DataLoaderConfig()
-    )
-
     prep = attr.ib(validator=optional(instance_of(PrepConfig)), default=None)
     train = attr.ib(validator=optional(instance_of(TrainConfig)), default=None)
     eval = attr.ib(validator=optional(instance_of(EvalConfig)), default=None)

src/vak/config/eval.py

@@ -99,10 +99,17 @@ class EvalConfig:
         a float value for ``min_segment_dur``.
         See the docstring of the transform for more details on
         these arguments and how they work.
+    transform_params: dict, optional
+        Parameters for data transform.
+        Passed as keyword arguments.
+        Optional, default is None.
+    dataset_params: dict, optional
+        Parameters for dataset.
+        Passed as keyword arguments.
+        Optional, default is None.
     """
     # required, external files
     checkpoint_path = attr.ib(converter=expanded_user_path)
-    labelmap_path = attr.ib(converter=expanded_user_path)
     output_dir = attr.ib(converter=expanded_user_path)
 
     # required, model / dataloader
@@ -118,6 +125,10 @@ class EvalConfig:
         default=None,
     )
 
+    # "optional" but actually required for frame classification models
+    # TODO: check model family in __post_init__ and raise ValueError if labelmap
+    # TODO: not specified for a frame classification model?
+    labelmap_path = attr.ib(converter=converters.optional(expanded_user_path), default=None)
     # optional, transform
     spect_scaler_path = attr.ib(
         converter=converters.optional(expanded_user_path),
@@ -133,3 +144,15 @@ class EvalConfig:
     # optional, data loader
     num_workers = attr.ib(validator=instance_of(int), default=2)
     device = attr.ib(validator=instance_of(str), default=device.get_default())
+
+    transform_params = attr.ib(
+        converter=converters.optional(dict),
+        validator=validators.optional(instance_of(dict)),
+        default=None,
+    )
+
+    dataset_params = attr.ib(
+        converter=converters.optional(dict),
+        validator=validators.optional(instance_of(dict)),
+        default=None,
+    )

src/vak/config/parse.py

@@ -4,7 +4,6 @@
 from toml.decoder import TomlDecodeError
 
 from .config import Config
-from .dataloader import DataLoaderConfig
 from .eval import EvalConfig
 from .learncurve import LearncurveConfig
 from .predict import PredictConfig
@@ -15,7 +14,6 @@
 
 
 SECTION_CLASSES = {
-    "DATALOADER": DataLoaderConfig,
     "EVAL": EvalConfig,
     "LEARNCURVE": LearncurveConfig,
     "PREDICT": PredictConfig,
@@ -25,10 +23,8 @@
 }
 
 REQUIRED_OPTIONS = {
-    "DATALOADER": None,
     "EVAL": [
         "checkpoint_path",
-        "labelmap_path",
         "output_dir",
         "model",
     ],
@@ -38,7 +34,6 @@
     ],
     "PREDICT": [
         "checkpoint_path",
-        "labelmap_path",
         "model",
     ],
     "PREP": [

src/vak/config/predict.py

@@ -68,6 +68,14 @@ class PredictConfig:
          spectrogram with `spect_path` filename `gy6or6_032312_081416.npz`,
          and the network is `TweetyNet`, then the net output file
          will be `gy6or6_032312_081416.tweetynet.output.npz`.
+    transform_params: dict, optional
+        Parameters for data transform.
+        Passed as keyword arguments.
+        Optional, default is None.
+    dataset_params: dict, optional
+        Parameters for dataset.
+        Passed as keyword arguments.
+        Optional, default is None.
     """
 
     # required, external files
@@ -109,3 +117,15 @@ class PredictConfig:
     )
     majority_vote = attr.ib(validator=instance_of(bool), default=True)
     save_net_outputs = attr.ib(validator=instance_of(bool), default=False)
+
+    transform_params = attr.ib(
+        converter=converters.optional(dict),
+        validator=validators.optional(instance_of(dict)),
+        default=None,
+    )
+
+    dataset_params = attr.ib(
+        converter=converters.optional(dict),
+        validator=validators.optional(instance_of(dict)),
+        default=None,
+    )

src/vak/config/train.py

@@ -118,3 +118,27 @@ class TrainConfig:
         converter=converters.optional(expanded_user_path),
         default=None,
     )
+
+    train_transform_params = attr.ib(
+        converter=converters.optional(dict),
+        validator=validators.optional(instance_of(dict)),
+        default=None,
+    )
+
+    train_dataset_params = attr.ib(
+        converter=converters.optional(dict),
+        validator=validators.optional(instance_of(dict)),
+        default=None,
+    )
+
+    val_transform_params = attr.ib(
+        converter=converters.optional(dict),
+        validator=validators.optional(instance_of(dict)),
+        default=None,
+    )
+
+    val_dataset_params = attr.ib(
+        converter=converters.optional(dict),
+        validator=validators.optional(instance_of(dict)),
+        default=None,
+    )

src/vak/config/valid.toml

@@ -33,9 +33,6 @@ freqbins_key = 'f'
 timebins_key = 't'
 audio_path_key = 'audio_path'
 
-[DATALOADER]
-window_size = 88
-
 [TRAIN]
 model = 'TweetyNet'
 root_results_dir = './tests/test_data/results/train'
@@ -52,6 +49,10 @@ patience = 4
 results_dir_made_by_main_script = '/some/path/to/learncurve/'
 checkpoint_path = '/home/user/results_181014_194418/TweetyNet/checkpoints/'
 spect_scaler_path = '/home/user/results_181014_194418/spect_scaler'
+train_transform_params = {'resize' = 128}
+train_dataset_params = {'window_size' = 80}
+val_transform_params = {'resize' = 128}
+val_dataset_params = {'window_size' = 80}
 
 [EVAL]
 dataset_path = 'tests/test_data/prep/learncurve/032312_prep_191224_225910.csv'
@@ -64,6 +65,8 @@ num_workers = 4
 device = 'cuda'
 spect_scaler_path = '/home/user/results_181014_194418/spect_scaler'
 post_tfm_kwargs = {'majority_vote' = true, 'min_segment_dur' = 0.01}
+transform_params = {'resize' = 128}
+dataset_params = {'window_size' = 80}
 
 [LEARNCURVE]
 model = 'TweetyNet'
@@ -80,7 +83,10 @@ results_dir_made_by_main_script = '/some/path/to/learncurve/'
 post_tfm_kwargs = {'majority_vote' = true, 'min_segment_dur' = 0.01}
 num_workers = 4
 device = 'cuda'
-
+train_transform_params = {'resize' = 128}
+train_dataset_params = {'window_size' = 80}
+val_transform_params = {'resize' = 128}
+val_dataset_params = {'window_size' = 80}
 
 [PREDICT]
 dataset_path = 'tests/test_data/prep/learncurve/032312_prep_191224_225910.csv'
@@ -96,3 +102,5 @@ spect_scaler_path = '/home/user/results_181014_194418/spect_scaler'
 min_segment_dur = 0.004
 majority_vote = false
 save_net_outputs = false
+transform_params = {'resize' = 128}
+dataset_params = {'window_size' = 80}

src/vak/datasets/__init__.py

@@ -1,8 +1,10 @@
 from . import (
     frame_classification,
+    parametric_umap
 )
 
 
 __all__ = [
+    "dimensionality_reduction",
     "frame_classification",
 ]

src/vak/datasets/frame_classification/frames_dataset.py

@@ -44,12 +44,15 @@ def __init__(
         dataset_df = dataset_df[dataset_df.split == split].copy()
         self.dataset_df = dataset_df
         self.frames_paths = self.dataset_df[constants.FRAMES_NPY_PATH_COL_NAME].values
-        self.frame_labels_paths = self.dataset_df[constants.FRAME_LABELS_NPY_PATH_COL_NAME].values
+        if split != 'predict':
+            self.frame_labels_paths = self.dataset_df[constants.FRAME_LABELS_NPY_PATH_COL_NAME].values
+        else:
+            self.frame_labels_paths = None
 
         if input_type == 'audio':
             self.source_paths = self.dataset_df['audio_path'].values
         elif input_type == 'spect':
-            self.source_paths = self.dataset_df['audio_path'].values
+            self.source_paths = self.dataset_df['spect_path'].values
         else:
             raise ValueError(
                 f"Invalid `input_type`: {input_type}. Must be one of {{'audio', 'spect'}}."
@@ -97,6 +100,18 @@ def from_dataset_path(
         split: str = "val",
         item_transform: Callable | None = None,
     ):
+        """
+
+        Parameters
+        ----------
+        dataset_path
+        split
+        item_transform
+
+        Returns
+        -------
+
+        """
         dataset_path = pathlib.Path(dataset_path)
         metadata = Metadata.from_dataset_path(dataset_path)
         frame_dur = metadata.frame_dur

src/vak/datasets/frame_classification/window_dataset.py

@@ -197,6 +197,21 @@ def from_dataset_path(
             transform: Callable | None = None,
             target_transform: Callable | None = None
     ):
+        """
+
+        Parameters
+        ----------
+        dataset_path
+        window_size
+        stride
+        split
+        transform
+        target_transform
+
+        Returns
+        -------
+
+        """
         dataset_path = pathlib.Path(dataset_path)
         metadata = Metadata.from_dataset_path(dataset_path)
         frame_dur = metadata.frame_dur

src/vak/datasets/parametric_umap/__init__.py

@@ -0,0 +1,8 @@
+from .metadata import Metadata
+from .parametric_umap import ParametricUMAPDataset
+
+
+__all__ = [
+    'Metadata',
+    'ParametricUMAPDataset'
+]