Rename Serializer to Output and move to flash.core.data.io.output (#…

…927)

Co-authored-by: Ananya Harsh Jha <ananya@pytorchlightning.ai>

CHANGELOG.md

@@ -10,8 +10,17 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Changed
 
+- Changed classes named `*Serializer` and properties / variables named `serializer` to be `*Output` and `output` respectively ([#927](https://github.com/PyTorchLightning/lightning-flash/pull/927))
+
+### Deprecated
+
+- Deprecated `flash.core.data.process.Serializer` in favour of `flash.core.data.io.output.Output` ([#927](https://github.com/PyTorchLightning/lightning-flash/pull/927))
+
+- Deprecated `Task.serializer` in favour of `Task.output` ([#927](https://github.com/PyTorchLightning/lightning-flash/pull/927))
+
 ### Fixed
 
+
 ## [0.5.2] - 2021-11-05
 
 ### Added

docs/source/api/core.rst

@@ -27,12 +27,12 @@ _________________________
     :template: classtemplate.rst
 
     ~flash.core.classification.Classes
-    ~flash.core.classification.ClassificationSerializer
+    ~flash.core.classification.ClassificationOutput
     ~flash.core.classification.ClassificationTask
     ~flash.core.classification.FiftyOneLabels
     ~flash.core.classification.Labels
     ~flash.core.classification.Logits
-    ~flash.core.classification.PredsClassificationSerializer
+    ~flash.core.classification.PredsClassificationOutput
     ~flash.core.classification.Probabilities
 
 flash.core.finetuning

docs/source/api/data.rst

@@ -7,6 +7,17 @@ flash.core.data
     :local:
     :backlinks: top
 
+flash.core.data.io.output
+_________________________
+
+.. autosummary::
+    :toctree: generated/
+    :nosignatures:
+    :template: classtemplate.rst
+
+    ~flash.core.data.io.output.Output
+    ~flash.core.data.io.output.OutputMapping
+
 flash.core.data.auto_dataset
 ____________________________
 
@@ -114,8 +125,6 @@ _______________________
     ~flash.core.data.process.Deserializer
     ~flash.core.data.process.Postprocess
     ~flash.core.data.process.Preprocess
-    ~flash.core.data.process.SerializerMapping
-    ~flash.core.data.process.Serializer
 
 flash.core.data.properties
 __________________________

docs/source/api/flash.rst

@@ -12,6 +12,6 @@ flash
     ~flash.core.data.callback.FlashCallback
     ~flash.core.data.process.Preprocess
     ~flash.core.data.process.Postprocess
-    ~flash.core.data.process.Serializer
+    ~flash.core.data.io.output.Output
     ~flash.core.model.Task
     ~flash.core.trainer.Trainer

docs/source/api/image.rst

@@ -45,7 +45,7 @@ ________________
     detection.data.FiftyOneParser
     detection.data.ObjectDetectionFiftyOneDataSource
     detection.data.ObjectDetectionPreprocess
-    detection.serialization.FiftyOneDetectionLabels
+    detection.output.FiftyOneDetectionLabels
 
 Keypoint Detection
 __________________
@@ -102,8 +102,8 @@ ____________
     segmentation.data.SemanticSegmentationFiftyOneDataSource
     segmentation.data.SemanticSegmentationDeserializer
     segmentation.model.SemanticSegmentationPostprocess
-    segmentation.serialization.FiftyOneSegmentationLabels
-    segmentation.serialization.SegmentationLabels
+    segmentation.output.FiftyOneSegmentationLabels
+    segmentation.output.SegmentationLabels
 
 .. autosummary::
     :toctree: generated/

docs/source/common/finetuning_example.rst

@@ -55,8 +55,8 @@ Once you've finetuned, use the model to predict:
 
 .. testcode:: finetune
 
-    # Serialize predictions as labels, automatically inferred from the training data in part 2.
-    model.serializer = Labels()
+    # Output predictions as labels, automatically inferred from the training data in part 2.
+    model.output = Labels()
 
     predictions = model.predict(
         [

docs/source/general/data.rst

@@ -26,7 +26,7 @@ Here are common terms you need to be familiar with:
    * - :class:`~flash.core.data.data_module.DataModule`
      - The :class:`~flash.core.data.data_module.DataModule` contains the datasets, transforms and dataloaders.
    * - :class:`~flash.core.data.data_pipeline.DataPipeline`
-     - The :class:`~flash.core.data.data_pipeline.DataPipeline` is Flash internal object to manage :class:`~flash.core.data.Deserializer`, :class:`~flash.core.data.data_source.DataSource`, :class:`~flash.core.data.process.Preprocess`, :class:`~flash.core.data.process.Postprocess`, and :class:`~flash.core.data.process.Serializer` objects.
+     - The :class:`~flash.core.data.data_pipeline.DataPipeline` is Flash internal object to manage :class:`~flash.core.data.Deserializer`, :class:`~flash.core.data.data_source.DataSource`, :class:`~flash.core.data.process.Preprocess`, :class:`~flash.core.data.process.Postprocess`, and :class:`~flash.core.data.io.output.Output` objects.
    * - :class:`~flash.core.data.data_source.DataSource`
      - The :class:`~flash.core.data.data_source.DataSource` provides :meth:`~flash.core.data.data_source.DataSource.load_data` and :meth:`~flash.core.data.data_source.DataSource.load_sample` hooks for creating data sets from metadata (such as folder names).
    * - :class:`~flash.core.data.process.Preprocess`
@@ -37,8 +37,8 @@ Here are common terms you need to be familiar with:
    * - :class:`~flash.core.data.process.Postprocess`
      - The :class:`~flash.core.data.process.Postprocess` provides a simple hook-based API to encapsulate your post-processing logic.
         The :class:`~flash.core.data.process.Postprocess` hooks cover from model outputs to predictions export.
-   * - :class:`~flash.core.data.process.Serializer`
-     - The :class:`~flash.core.data.process.Serializer` provides a single :meth:`~flash.core.data.process.Serializer.serialize` method that is used to convert model outputs (after the :class:`~flash.core.data.process.Postprocess`) to the desired output format during prediction.
+   * - :class:`~flash.core.data.io.output.Output`
+     - The :class:`~flash.core.data.io.output.Output` provides a single :meth:`~flash.core.data.io.output.Output.serialize` method that is used to convert model outputs (after the :class:`~flash.core.data.process.Postprocess`) to the desired output format during prediction.
 
 
 *******************************************
@@ -59,7 +59,7 @@ Usually, extra processing logic should be added to bridge the gap between traini
 
 The :class:`~flash.core.data.data_source.DataSource` class can be used to generate data sets from multiple sources (e.g. folders, numpy, etc.), that can then all be transformed in the same way.
 The :class:`~flash.core.data.process.Preprocess` and :class:`~flash.core.data.process.Postprocess` classes can be used to manage the preprocessing and postprocessing transforms.
-The :class:`~flash.core.data.process.Serializer` class provides the logic for converting :class:`~flash.core.data.process.Postprocess` outputs to the desired predict format (e.g. classes, labels, probabilities, etc.).
+The :class:`~flash.core.data.io.output.Output` class provides the logic for converting :class:`~flash.core.data.process.Postprocess` outputs to the desired predict format (e.g. classes, labels, probabilities, etc.).
 
 By providing a series of hooks that can be overridden with custom data processing logic (or just targeted with transforms),
 Flash gives the user much more granular control over their data processing flow.
@@ -383,18 +383,18 @@ Example::
     predictions = lightning_module(data)
 
 
-Postprocess and Serializer
+Postprocess and Output
 __________________________
 
 
 Once the predictions have been generated by the Flash :class:`~flash.core.model.Task`, the Flash
 :class:`~flash.core.data.data_pipeline.DataPipeline` will execute the :class:`~flash.core.data.process.Postprocess` hooks and the
-:class:`~flash.core.data.process.Serializer` behind the scenes.
+:class:`~flash.core.data.io.output.Output` behind the scenes.
 
 First, the :meth:`~flash.core.data.process.Postprocess.per_batch_transform` hooks will be applied on the batch predictions.
 Then, the :meth:`~flash.core.data.process.Postprocess.uncollate` will split the batch into individual predictions.
 Next, the :meth:`~flash.core.data.process.Postprocess.per_sample_transform` will be applied on each prediction.
-Finally, the :meth:`~flash.core.data.process.Serializer.serialize` method will be called to serialize the predictions.
+Finally, the :meth:`~flash.core.data.io.output.Output.serialize` method will be called to serialize the predictions.
 
 .. note:: The transform can be applied either on device or ``CPU``.
 

docs/source/general/predictions.rst

@@ -57,8 +57,8 @@ Predict on a csv file
 Serializing predictions
 =======================
 
-To change how predictions are serialized you can attach a :class:`~flash.core.data.process.Serializer` to your
-:class:`~flash.core.model.Task`. For example, you can choose to serialize outputs as probabilities (for more options see the API
+To change the output format of predictions you can attach an :class:`~flash.core.data.io.output.Output` to your
+:class:`~flash.core.model.Task`. For example, you can choose to output probabilities (for more options see the API
 reference below).
 
 
@@ -77,8 +77,8 @@ reference below).
         "https://flash-weights.s3.amazonaws.com/0.5.2/image_classification_model.pt"
     )
 
-    # 3. Attach the Serializer
-    model.serializer = Probabilities()
+    # 3. Attach the Output
+    model.output = Probabilities()
 
     # 4. Predict whether the image contains an ant or a bee
     predictions = model.predict("data/hymenoptera_data/val/bees/65038344_52a45d090d.jpg")

docs/source/integrations/fiftyone.rst

@@ -47,8 +47,8 @@ semantic segmentation tasks. Doing so is as easy as updating your model to use
 one of the following serializers:
 
 * :class:`FiftyOneLabels(return_filepath=True)<flash.core.classification.FiftyOneLabels>`
-* :class:`FiftyOneSegmentationLabels(return_filepath=True)<flash.image.segmentation.serialization.FiftyOneSegmentationLabels>`
-* :class:`FiftyOneDetectionLabels(return_filepath=True)<flash.image.detection.serialization.FiftyOneDetectionLabels>`
+* :class:`FiftyOneSegmentationLabels(return_filepath=True)<flash.image.segmentation.output.FiftyOneSegmentationLabels>`
+* :class:`FiftyOneDetectionLabels(return_filepath=True)<flash.image.detection.output.FiftyOneDetectionLabels>`
 
 The :func:`~flash.core.integrations.fiftyone.visualize` function then lets you visualize
 your predictions in the

docs/source/template/optional.rst

@@ -20,21 +20,21 @@ Here's how we create our transforms in the :class:`~flash.image.classification.d
     :language: python
     :pyobject: ImageClassificationPreprocess.default_transforms
 
-Add output serializers to your Task
-======================================
+Add outputs to your Task
+========================
 
 We recommend that you do most of the heavy lifting in the :class:`~flash.core.data.process.Postprocess`.
 Specifically, it should include any formatting and transforms that should always be applied to the predictions.
-If you want to support different use cases that require different prediction formats, you should add some :class:`~flash.core.data.process.Serializer` implementations in a ``serialization.py`` file.
+If you want to support different use cases that require different prediction formats, you should add some :class:`~flash.core.data.io.output.Output` implementations in an ``output.py`` file.
 
 Some good examples are in `flash/core/classification.py <https://github.com/PyTorchLightning/lightning-flash/blob/master/flash/core/classification.py>`_.
-Here's the :class:`~flash.core.classification.Classes` :class:`~flash.core.data.process.Serializer`:
+Here's the :class:`~flash.core.classification.Classes` :class:`~flash.core.data.io.output.Output`:
 
 .. literalinclude:: ../../../flash/core/classification.py
     :language: python
     :pyobject: Classes
 
-Alternatively, here's the :class:`~flash.core.classification.Logits` :class:`~flash.core.data.process.Serializer`:
+Alternatively, here's the :class:`~flash.core.classification.Logits` :class:`~flash.core.data.io.output.Output`:
 
 .. literalinclude:: ../../../flash/core/classification.py
     :language: python

flash/__init__.py

@@ -24,6 +24,7 @@
     from flash.core.data.data_source import DataSource
     from flash.core.data.datasets import FlashDataset, FlashIterableDataset
     from flash.core.data.input_transform import InputTransform
+    from flash.core.data.io.output import Output
     from flash.core.data.process import Postprocess, Preprocess, Serializer
     from flash.core.model import Task  # noqa: E402
     from flash.core.trainer import Trainer  # noqa: E402
@@ -44,9 +45,10 @@
         "FlashCallback",
         "FlashDataset",
         "FlashIterableDataset",
-        "Preprocess",
         "InputTransform",
+        "Output",
         "Postprocess",
+        "Preprocess",
         "Serializer",
         "Task",
         "Trainer",

flash/audio/speech_recognition/model.py

@@ -25,7 +25,7 @@
 from flash.core.model import Task
 from flash.core.registry import FlashRegistry
 from flash.core.utilities.imports import _AUDIO_AVAILABLE
-from flash.core.utilities.types import LR_SCHEDULER_TYPE, OPTIMIZER_TYPE, SERIALIZER_TYPE
+from flash.core.utilities.types import LR_SCHEDULER_TYPE, OPTIMIZER_TYPE, OUTPUT_TYPE
 
 if _AUDIO_AVAILABLE:
     from transformers import Wav2Vec2Processor
@@ -41,7 +41,7 @@ class SpeechRecognition(Task):
         learning_rate: Learning rate to use for training, defaults to ``1e-5``.
         optimizer: Optimizer to use for training.
         lr_scheduler: The LR scheduler to use during training.
-        serializer: The :class:`~flash.core.data.process.Serializer` to use when serializing prediction outputs.
+        output: The :class:`~flash.core.data.io.output.Output` to use when serializing prediction outputs.
     """
 
     backbones: FlashRegistry = SPEECH_RECOGNITION_BACKBONES
@@ -54,7 +54,7 @@ def __init__(
         optimizer: OPTIMIZER_TYPE = "Adam",
         lr_scheduler: LR_SCHEDULER_TYPE = None,
         learning_rate: float = 1e-5,
-        serializer: SERIALIZER_TYPE = None,
+        output: OUTPUT_TYPE = None,
     ):
         os.environ["TOKENIZERS_PARALLELISM"] = "TRUE"
         # disable HF thousand warnings
@@ -68,7 +68,7 @@ def __init__(
             optimizer=optimizer,
             lr_scheduler=lr_scheduler,
             learning_rate=learning_rate,
-            serializer=serializer,
+            output=output,
         )
 
         self.save_hyperparameters()