Use new data format

components/caption_images/fondant_component.yaml

@@ -5,16 +5,12 @@ tags:
   - Image processing
 
 consumes:
-  images:
-    fields:
-      data:
-        type: binary
+  images_data:
+    type: binary
 
 produces:
-  captions:
-    fields:
-      text:
-        type: utf8
+  captions_text:
+    type: utf8
 
 args:
   model_id:

components/caption_images/src/main.py

@@ -90,7 +90,7 @@ def __init__(
         self.max_new_tokens = max_new_tokens
 
     def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
-        images = dataframe["images"]["data"]
+        images = dataframe["images_data"]
 
         results: t.List[pd.Series] = []
         for batch in np.split(
@@ -112,4 +112,4 @@ def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
                 ).T
                 results.append(captions)
 
-        return pd.concat(results).to_frame(name=("captions", "text"))
+        return pd.concat(results).to_frame(name=("captions_text"))

components/chunk_text/fondant_component.yaml

@@ -10,18 +10,14 @@ tags:
   - Text processing
 
 consumes:
-  text:
-    fields:
-      data:
-        type: string
+  text_data:
+    type: string
 
 produces:
-  text:
-    fields:
-      data:
-        type: string
-      original_document_id:
-        type: string
+  text_data:
+    type: string
+  text_original_document_id:
+    type: string
 
 args:
   chunk_size:

components/chunk_text/src/main.py

@@ -38,7 +38,7 @@ def __init__(
     def chunk_text(self, row) -> t.List[t.Tuple]:
         # Multi-index df has id under the name attribute
         doc_id = row.name
-        text_data = row[("text", "data")]
+        text_data = row[("text_data")]
         docs = self.text_splitter.create_documents([text_data])
         return [
             (doc_id, f"{doc_id}_{chunk_id}", chunk.page_content)
@@ -63,9 +63,4 @@ def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
         )
         results_df = results_df.set_index("id")
 
-        # Set multi-index column for the expected subset and field
-        results_df.columns = pd.MultiIndex.from_product(
-            [["text"], results_df.columns],
-        )
-
         return results_df

components/chunk_text/tests/chunk_text_test.py

@@ -3,11 +3,11 @@
 from src.main import ChunkTextComponent
 
 
-def test_transform():
+def test_transform__():
     """Test chunk component method."""
     input_dataframe = pd.DataFrame(
         {
-            ("text", "data"): [
+            ("text_data"): [
                 "Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Aenean commodo",
                 "ligula eget dolor. Aenean massa. Cum sociis natoque penatibus et magnis dis",
                 "parturient montes, nascetur ridiculus mus. Donec quam felis, ultricies nec,",
@@ -25,8 +25,8 @@ def test_transform():
 
     expected_output_dataframe = pd.DataFrame(
         {
-            ("text", "original_document_id"): ["a", "a", "a", "b", "b", "c", "c"],
-            ("text", "data"): [
+            ("text_original_document_id"): ["a", "a", "a", "b", "b", "c", "c"],
+            ("text_data"): [
                 "Lorem ipsum dolor sit amet, consectetuer",
                 "amet, consectetuer adipiscing elit. Aenean",
                 "elit. Aenean commodo",

components/download_images/fondant_component.yaml

@@ -13,21 +13,17 @@ tags:
   - Image processing
 
 consumes:
-  images:
-    fields:
-      url:
-        type: string
+  images_url:
+    type: string
 
 produces:
-  images:
-    fields:
-      data:
-        type: binary
-      width:
-        type: int32
-      height:
-        type: int32
-    additionalFields: false
+  images_data:
+    type: binary
+  images_width:
+    type: int32
+  images_height:
+    type: int32
+#    additionalFields: false
 
 args:
   timeout:
@@ -53,7 +49,7 @@ args:
     description: Resize mode to use. One of "no", "keep_ratio", "center_crop", "border".
     type: str
     default: 'border'
-  resize_only_if_bigger: 
+  resize_only_if_bigger:
     description: If True, resize only if image is bigger than image_size.
     type: bool
     default: False

components/download_images/src/main.py

@@ -119,7 +119,7 @@ async def download_dataframe() -> None:
             images = await asyncio.gather(
                 *[
                     self.download_and_resize_image(id_, url, semaphore=semaphore)
-                    for id_, url in zip(dataframe.index, dataframe["images"]["url"])
+                    for id_, url in zip(dataframe.index, dataframe["images_url"])
                 ],
             )
             results.extend(images)
@@ -134,8 +134,5 @@ async def download_dataframe() -> None:
 
         results_df = results_df.dropna()
         results_df = results_df.set_index("id", drop=True)
-        results_df.columns = pd.MultiIndex.from_product(
-            [["images"], results_df.columns],
-        )
 
         return results_df

components/download_images/tests/test_component.py

@@ -45,7 +45,7 @@ def test_transform(respx_mock):
 
     input_dataframe = pd.DataFrame(
         {
-            ("images", "url"): urls,
+            "images_url": urls,
         },
         index=pd.Index(ids, name="id"),
     )
@@ -55,9 +55,9 @@ def test_transform(respx_mock):
     resized_images = [component.resizer(io.BytesIO(image))[0] for image in images]
     expected_dataframe = pd.DataFrame(
         {
-            ("images", "data"): resized_images,
-            ("images", "width"): [image_size] * len(ids),
-            ("images", "height"): [image_size] * len(ids),
+            "images_data": resized_images,
+            "images_width": [image_size] * len(ids),
+            "images_height": [image_size] * len(ids),
         },
         index=pd.Index(ids, name="id"),
     )

components/embed_images/fondant_component.yaml

@@ -2,21 +2,17 @@ name: Embed images
 description: Component that generates CLIP embeddings from images
 image: fndnt/embed_images:dev
 tags:
-   - Image processing
+  - Image processing
 
 consumes:
-  images:
-      fields:
-        data:
-          type: binary
+  images_data:
+    type: binary
 
 produces:
-  embeddings:
-    fields:
-      data:
-        type: array
-        items:
-          type: float32
+  embeddings_data:
+    type: array
+    items:
+      type: float32
 
 args:
   model_id:

components/embed_images/src/main.py

@@ -90,7 +90,7 @@ def __init__(
         self.batch_size = batch_size
 
     def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
-        images = dataframe["images"]["data"]
+        images = dataframe["images_data"]
 
         results: t.List[pd.Series] = []
         for batch in np.split(
@@ -110,4 +110,4 @@ def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
                 ).T
                 results.append(embeddings)
 
-        return pd.concat(results).to_frame(name=("embeddings", "data"))
+        return pd.concat(results).to_frame(name=("embeddings_data"))

components/embed_text/fondant_component.yaml

@@ -5,21 +5,17 @@ tags:
   - Text processing
 
 consumes:
-  text:
-    fields:
-      data:
-        type: string
+  text_data:
+    type: string
 
 produces:
-  text:
-    fields:
-      data:
-        type: string
-      embedding:
-        type: array
-        items:
-          type: float32
-
+  text_data:
+    type: string
+  text_embedding:
+    type: array
+    items:
+      type: float32
+
 args:
   model_provider:
     description: |
@@ -40,12 +36,12 @@ args:
       Pass only the keys required by the model provider or conveniently pass all keys you will ever need. 
       Pay attention how to name the dictionary keys so that they can be used by the model provider.
     type: dict
-    default: {}
+    default: { }
   auth_kwargs:
     description: |
       Additional keyword arguments required for api initialization/authentication.
     type: dict
-    default: {}
+    default: { }
 
 
 

components/embed_text/src/main.py

@@ -65,7 +65,7 @@ def get_embeddings_vectors(self, texts):
         return self.embedding_model.embed_documents(texts.tolist())
 
     def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
-        dataframe[("text", "embedding")] = self.get_embeddings_vectors(
-            dataframe[("text", "data")],
+        dataframe["text_embedding"] = self.get_embeddings_vectors(
+            dataframe["text_data"],
         )
         return dataframe

components/embedding_based_laion_retrieval/fondant_component.yaml

@@ -7,18 +7,14 @@ tags:
   - Data retrieval
 
 consumes:
-  embeddings:
-    fields:
-      data:
-        type: array
-        items:
-          type: float32
+  embeddings_data:
+    type: array
+    items:
+      type: float32
 
 produces:
-  images:
-    fields:
-      url:
-        type: string
+  images_url:
+    type: string
   additionalSubsets: false
 
 args:

components/embedding_based_laion_retrieval/src/main.py

@@ -58,18 +58,18 @@ async def async_query():
                             embedding_input=embedding.tolist(),
                         ),
                     )
-                    for embedding in dataframe["embeddings"]["data"]
+                    for embedding in dataframe["embeddings_data"]
                 ]
                 for response in await asyncio.gather(*futures):
                     results.extend(response)
 
         loop.run_until_complete(async_query())
 
-        results_df = pd.DataFrame(results)[["id", "url"]]
+        results_df = pd.DataFrame(results)["id", "url"]
         results_df = results_df.set_index("id")
 
         # Cast the index to string
         results_df.index = results_df.index.astype(str)
-        results_df.columns = [["images"], ["url"]]
+        results_df.columns = ["images_url"]
 
         return results_df

components/filter_image_resolution/fondant_component.yaml

@@ -5,12 +5,10 @@ tags:
   - Image processing
 
 consumes:
-  images:
-    fields:
-      width:
-        type: int32
-      height:
-        type: int32
+  images_width:
+    type: int32
+  images_height:
+    type: int32
 
 args:
   min_image_dim:

components/filter_image_resolution/src/main.py

@@ -23,8 +23,8 @@ def __init__(self, *_, min_image_dim: int, max_aspect_ratio: float) -> None:
         self.max_aspect_ratio = max_aspect_ratio
 
     def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
-        width = dataframe["images"]["width"]
-        height = dataframe["images"]["height"]
+        width = dataframe["images_width"]
+        height = dataframe["images_height"]
         min_image_dim = np.minimum(width, height)
         max_image_dim = np.maximum(width, height)
         aspect_ratio = max_image_dim / min_image_dim

components/filter_text_length/fondant_component.yaml

@@ -5,10 +5,8 @@ tags:
   - Text processing
 
 consumes:
-  text:
-    fields:
-      data:
-        type: string
+  text_data:
+    type: string
 
 args:
   min_characters_length:

components/filter_text_length/src/main.py

@@ -23,10 +23,10 @@ def __init__(self, *_, min_characters_length: int, min_words_length: int):
 
     def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
         """Filter out text based on their length."""
-        caption_num_words = dataframe["text"]["data"].apply(
+        caption_num_words = dataframe["text_data"].apply(
             lambda x: len(fasttext.tokenize(x)),
         )
-        caption_num_chars = dataframe["text"]["data"].apply(len)
+        caption_num_chars = dataframe["text_data"].apply(len)
 
         mask = (caption_num_words >= self.min_words_length) & (
             caption_num_chars >= self.min_characters_length

components/filter_text_length/tests/text_length_filter_test.py

@@ -24,6 +24,6 @@ def test_run_component_test():
     # Then: dataframe only contains one row
     assert len(dataframe) == 1
     assert (
-        dataframe.loc[2]["text"]["data"]
+        dataframe.loc[2]["text_data"]
         == "This a valid sentence which should be still there"
     )