Onboard text classification inputs to new object (#1022)

.gitignore

@@ -173,4 +173,5 @@ mini-imagenet*
 
 docs/source/_static/images/course_UvA-DL
 docs/source/_static/images/lightning_examples
+docs/source/_static/images/flash_tutorials
 docs/source/notebooks

CHANGELOG.md

@@ -78,6 +78,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Removed `Task.predict`, use `Trainer.predict` instead ([#1030](https://github.com/PyTorchLightning/lightning-flash/pull/1030))
 
+- Removed the `backbone` argument from `TextClassificationData`, it is now sufficient to only provide a `backbone` argument to the `TextClassifier` ([#1022](https://github.com/PyTorchLightning/lightning-flash/pull/1022))
+
 ## [0.5.2] - 2021-11-05
 
 ### Added

docs/source/api/text.rst

@@ -20,16 +20,14 @@ ______________
     ~classification.model.TextClassifier
     ~classification.data.TextClassificationData
 
-    classification.data.TextClassificationOutputTransform
     classification.data.TextClassificationInputTransform
     classification.data.TextDeserializer
-    classification.data.TextInput
-    classification.data.TextCSVInput
-    classification.data.TextJSONInput
-    classification.data.TextDataFrameInput
-    classification.data.TextParquetInput
-    classification.data.TextHuggingFaceDatasetInput
-    classification.data.TextListInput
+    classification.data.TextClassificationInput
+    classification.data.TextClassificationCSVInput
+    classification.data.TextClassificationJSONInput
+    classification.data.TextClassificationDataFrameInput
+    classification.data.TextClassificationParquetInput
+    classification.data.TextClassificationListInput
 
 Question Answering
 __________________

docs/source/template/data.rst

@@ -201,11 +201,11 @@ OutputTransform
 
 :class:`~flash.core.data.io.output_transform.OutputTransform` contains any transforms that need to be applied *after* the model.
 You may want to use it for: converting tokens back into text, applying an inverse normalization to an output image, resizing a generated image back to the size of the input, etc.
-As an example, here's the :class:`~text.classification.data.TextClassificationOutputTransform` which gets the logits from a ``SequenceClassifierOutput``:
+As an example, here's the :class:`~text.seq2seq.core.data.Seq2SeqOutputTransform` which decodes tokenized model outputs:
 
-.. literalinclude:: ../../../flash/text/classification/data.py
+.. literalinclude:: ../../../flash/text/seq2seq/core/data.py
     :language: python
-    :pyobject: TextClassificationOutputTransform
+    :pyobject: Seq2SeqOutputTransform
 
 In your :class:`~flash.core.data.io.input.Input` or :class:`~flash.core.data.io.input_transform.InputTransform`, you can add metadata to the batch using the :attr:`~flash.core.data.io.input.DataKeys.METADATA` key.
 Your :class:`~flash.core.data.io.output_transform.OutputTransform` can then use this metadata in its transforms.

flash/core/data/io/input.py

@@ -154,7 +154,7 @@ class InputFormat(LightningEnum):
     JSON = "json"
     PARQUET = "parquet"
     DATASETS = "datasets"
-    HUGGINGFACE_DATASET = "hf_dataset"
+    HUGGINGFACE_DATASET = "hf_datasets"
     FIFTYONE = "fiftyone"
     DATAFRAME = "data_frame"
     LISTS = "lists"

flash/core/data/splits.py

@@ -4,6 +4,7 @@
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
 from torch.utils.data import Dataset
 
+import flash
 from flash.core.data.properties import Properties
 
 
@@ -26,7 +27,11 @@ class SplitDataset(Properties, Dataset):
     def __init__(self, dataset: Any, indices: List[int] = None, use_duplicated_indices: bool = False) -> None:
         kwargs = {}
         if isinstance(dataset, Properties):
-            kwargs = {"running_stage": dataset._running_stage, "state": dataset._state}
+            kwargs = dict(
+                running_stage=dataset._running_stage,
+                data_pipeline_state=dataset._data_pipeline_state,
+                state=dataset._state,
+            )
         super().__init__(**kwargs)
 
         if indices is None:
@@ -45,6 +50,11 @@ def __init__(self, dataset: Any, indices: List[int] = None, use_duplicated_indic
         self.dataset = dataset
         self.indices = indices
 
+    def attach_data_pipeline_state(self, data_pipeline_state: "flash.core.data.data_pipeline.DataPipelineState"):
+        super().attach_data_pipeline_state(data_pipeline_state)
+        if isinstance(self.dataset, Properties):
+            self.dataset.attach_data_pipeline_state(data_pipeline_state)
+
     def __getattr__(self, key: str):
         if key != "dataset":
             return getattr(self.dataset, key)

flash/core/integrations/labelstudio/input.py

@@ -16,15 +16,12 @@
 from flash.core.data.io.input_base import Input, IterableInput
 from flash.core.data.properties import ProcessState, Properties
 from flash.core.data.utils import image_default_loader
-from flash.core.utilities.imports import _PYTORCHVIDEO_AVAILABLE, _TEXT_AVAILABLE
+from flash.core.utilities.imports import _PYTORCHVIDEO_AVAILABLE
 from flash.core.utilities.stages import RunningStage
 
 if _PYTORCHVIDEO_AVAILABLE:
     from pytorchvideo.data.clip_sampling import make_clip_sampler
 
-if _TEXT_AVAILABLE:
-    from transformers import AutoTokenizer
-
 
 @dataclass(unsafe_hash=True, frozen=True)
 class LabelStudioState(ProcessState):
@@ -277,11 +274,8 @@ class LabelStudioTextClassificationInput(LabelStudioInput):
     Export data should point to text data
     """
 
-    def __init__(self, *args, backbone=None, max_length=128, **kwargs):
-        if backbone:
-            self.backbone = backbone
-            self.tokenizer = AutoTokenizer.from_pretrained(backbone, use_fast=True)
-            self.max_length = max_length
+    def __init__(self, *args, max_length=128, **kwargs):
+        self.max_length = max_length
         super().__init__(*args, **kwargs)
 
     def load_sample(self, sample: Mapping[str, Any] = None) -> Any:
@@ -291,17 +285,17 @@ def load_sample(self, sample: Mapping[str, Any] = None) -> Any:
 
         assert self.state
 
-        if self.backbone:
-            data = ""
-            for key in sample.get("data"):
-                data += sample.get("data").get(key)
-            tokenized_data = self.tokenizer(data, max_length=self.max_length, truncation=True, padding="max_length")
-            for key in tokenized_data:
-                tokenized_data[key] = torch.tensor(tokenized_data[key])
-            tokenized_data["labels"] = _get_labels_from_sample(sample["label"], self.state.classes)
-            # separate text data type block
-            result = tokenized_data
-        return result
+        data = ""
+        for key in sample.get("data"):
+            data += sample.get("data").get(key)
+        tokenized_data = self.get_state(flash.text.classification.model.TextClassificationBackboneState).tokenizer(
+            data, max_length=self.max_length, truncation=True, padding="max_length"
+        )
+        for key in tokenized_data:
+            tokenized_data[key] = torch.tensor(tokenized_data[key])
+        tokenized_data["labels"] = _get_labels_from_sample(sample["label"], self.state.classes)
+        # separate text data type block
+        return tokenized_data
 
 
 class LabelStudioVideoClassificationInput(LabelStudioIterableInput):

flash/image/classification/data.py

@@ -383,10 +383,10 @@ def from_data_frame(
         predict_data = (predict_data_frame, input_field, predict_images_root, predict_resolver)
 
         return cls(
-            ImageClassificationCSVInput(RunningStage.TRAINING, *train_data, **dataset_kwargs),
-            ImageClassificationCSVInput(RunningStage.VALIDATING, *val_data, **dataset_kwargs),
-            ImageClassificationCSVInput(RunningStage.TESTING, *test_data, **dataset_kwargs),
-            ImageClassificationCSVInput(RunningStage.PREDICTING, *predict_data, **dataset_kwargs),
+            ImageClassificationDataFrameInput(RunningStage.TRAINING, *train_data, **dataset_kwargs),
+            ImageClassificationDataFrameInput(RunningStage.VALIDATING, *val_data, **dataset_kwargs),
+            ImageClassificationDataFrameInput(RunningStage.TESTING, *test_data, **dataset_kwargs),
+            ImageClassificationDataFrameInput(RunningStage.PREDICTING, *predict_data, **dataset_kwargs),
             input_transform=cls.input_transform_cls(
                 train_transform,
                 val_transform,

flash/text/classification/cli.py

@@ -20,10 +20,7 @@
 
 
 def from_imdb(
-    backbone: str = "prajjwal1/bert-medium",
-    batch_size: int = 4,
-    num_workers: int = 0,
-    **input_transform_kwargs,
+    **data_module_kwargs,
 ) -> TextClassificationData:
     """Downloads and loads the IMDB sentiment classification data set."""
     download_data("https://pl-flash-data.s3.amazonaws.com/imdb.zip", "./data/")
@@ -32,31 +29,22 @@ def from_imdb(
         "sentiment",
         train_file="data/imdb/train.csv",
         val_file="data/imdb/valid.csv",
-        backbone=backbone,
-        batch_size=batch_size,
-        num_workers=num_workers,
-        **input_transform_kwargs,
+        **data_module_kwargs,
     )
 
 
 def from_toxic(
-    backbone: str = "unitary/toxic-bert",
     val_split: float = 0.1,
-    batch_size: int = 4,
-    num_workers: int = 0,
-    **input_transform_kwargs,
+    **data_module_kwargs,
 ) -> TextClassificationData:
     """Downloads and loads the Jigsaw toxic comments data set."""
     download_data("https://pl-flash-data.s3.amazonaws.com/jigsaw_toxic_comments.zip", "./data")
     return TextClassificationData.from_csv(
         "comment_text",
         ["toxic", "severe_toxic", "obscene", "threat", "insult", "identity_hate"],
         train_file="data/jigsaw_toxic_comments/train.csv",
-        backbone=backbone,
         val_split=val_split,
-        batch_size=batch_size,
-        num_workers=num_workers,
-        **input_transform_kwargs,
+        **data_module_kwargs,
     )
 
 
@@ -70,8 +58,7 @@ def text_classification():
         default_arguments={
             "trainer.max_epochs": 3,
         },
-        datamodule_attributes={"num_classes", "multi_label", "backbone"},
-        legacy=True,
+        datamodule_attributes={"num_classes", "multi_label"},
     )
 
     cli.trainer.save_checkpoint("text_classification_model.pt")