Don't use from_registry for generic components

README.md

@@ -174,7 +174,7 @@ def build_pipeline():
     pipeline.add_op(load_from_hub_op)
 
     custom_op = ComponentOp(
-        component_spec_path="components/custom_component/fondant_component.yaml",
+        component_dir="components/custom_component",
         arguments={
             "min_width": 600,
             "min_height": 600,

docs/component_spec.md

@@ -143,7 +143,7 @@ If an argument is not explicitly provided, the default value will be used instea
 from fondant.pipeline import ComponentOp
 
 custom_op = ComponentOp(
-    component_spec_path="components/custom_component/fondant_component.yaml",
+    component_dir="components/custom_component",
     arguments={
         "custom_argument": "foo"
     },

docs/components.md

@@ -0,0 +1,148 @@
+# Components
+
+Fondant makes it easy to build data preparation pipelines leveraging reusable components. Fondant
+provides a lot of components out of the box
+([overview](https://github.com/ml6team/fondant/tree/main/components)), but you can also define your
+own custom components.
+
+## The anatomy of a component
+
+A component is completely defined by its [component specification](component_spec.md) and a 
+docker image. The specification defines the docker image fondant should run to execute the 
+component, which data it consumes and produces, and which arguments it takes.
+
+## Component types
+
+We can distinguish three different types of components:
+
+- **Reusable components** can be used out of the box and can be loaded from the fondant 
+  component registry
+- **Custom components** are completely defined and implemented by the user
+- **Generic components** leverage a reusable implementation, but require a custom component 
+  specification
+
+### Reusable components
+
+Reusable components are completely defined and implemented by fondant. You can easily add them 
+to your pipeline by creating an operation using `ComponentOp.from_registry()`.
+
+```python
+from fondant.pipeline import ComponentOp
+
+component_op = ComponentOp.from_registry(
+  name="reusable_component",
+  arguments={
+    "arg": "value"
+  }
+)
+```
+
+??? "fondant.pipeline.ComponentOp.from_registry"
+
+    ::: fondant.pipeline.ComponentOp.from_registry
+        handler: python
+        options:
+          show_source: false
+
+You can find an overview of the reusable components offered by fondant 
+[here](https://github.com/ml6team/fondant/tree/main/components). Check their 
+`fondant_component.yaml` file for information on which arguments they accept and which data they 
+consume and produce.
+
+### Custom components
+
+To define your own custom component, you can build your code into a docker image and write an 
+accompanying component specification that refers to it.
+
+A typical file structure for a custom component looks like this:
+```
+|- components
+|  |- custom_component
+|     |- src
+|     |  |- main.py
+|     |- Dockerfile
+|     |- fondant_component.yaml
+|- pipeline.py
+```
+
+The `Dockerfile` is used to build the code into a docker image, which is then referred to in the 
+`fondant_component.yaml`. 
+
+```yaml title="components/custom_component/fondant_component.yaml"
+name: Custom component
+description: This is a custom component
+image: custom_component:latest
+```
+
+You can add a custom component to your pipeline by creating a `ComponentOp` and passing in the path 
+to the directory containing your `fondant_component.yaml`.
+
+```python title="pipeline.py"
+from fondant.pipeline import ComponentOp
+
+component_op = ComponentOp(
+  component_dir="components/custom_component",
+  arguments={
+    "arg": "value"
+  }
+)
+```
+
+??? "fondant.pipeline.ComponentOp"
+
+    ::: fondant.pipeline.ComponentOp
+        handler: python
+        options:
+          members: []
+          show_source: false
+
+See our [best practices on creating a custom component](custom_component.md).
+
+### Generic components
+
+A generic component is a component leveraging a reusable docker image, but requiring a custom 
+`fondant_component.yaml` specification.
+
+Since a generic component only requires a custom `fondant_component.yaml`, its file structure 
+looks like this:
+```
+|- components
+|  |- generic_component
+|     |- fondant_component.yaml
+|- pipeline.py
+```
+
+The `fondant_component.yaml` refers to the reusable image it leverages:
+
+```yaml title="components/generic_component/fondant_component.yaml"
+name: Generic component
+description: This is a generic component
+image: reusable_component:latest
+```
+
+You can add a generic component to your pipeline by creating a `ComponentOp` and passing in the path
+to the directory containing your custom `fondant_component.yaml`.
+
+```python title="pipeline.py"
+from fondant.pipeline import ComponentOp
+
+component_op = ComponentOp(
+  component_dir="components/generic_component",
+  arguments={
+    "arg": "value"
+  }
+)
+```
+
+??? "fondant.pipeline.ComponentOp"
+
+    ::: fondant.pipeline.ComponentOp
+        handler: python
+        options:
+          members: []
+          show_source: false
+
+An example of a generic component is the 
+[`load_from_hf_hub`](https://github.com/ml6team/fondant/tree/main/components/load_from_hf_hub) 
+components. It can read any dataset from the HuggingFace hub, but it requires the user to define 
+the schema of the produced dataset in a custom `fondant_component.yaml` specification.

docs/getting_started.md

@@ -48,9 +48,8 @@ Now that we have a pipeline, we can add components to it. Components are the bui
 Let's add a reusable component to our pipeline. We will use the `load_from_hf_hub` component to read data from huggingface. Add the following code to your `pipeline.py` file:
 
 ```Python
-load_from_hf_hub = ComponentOp.from_registry(
-    name='load_from_hf_hub',
-    component_spec_path='components/load_from_hf_hub/fondant_component.yml',
+load_from_hf_hub = ComponentOp(
+    component_dir='components/load_from_hf_hub',
     arguments={
         'dataset_name': 'huggan/pokemon',
         'n_rows_to_load': 100,
@@ -278,7 +277,7 @@ With our component complete we can now add it to our pipeline definition (`pipel
 
 ```python
 extract_resolution = ComponentOp(
-    component_spec_path='components/extract_resolution/fondant_component.yml', 
+    component_dir='components/extract_resolution', 
 )
 
 my_pipeline.add_op(load_from_hf_hub) # this line was already there

docs/pipeline.md

@@ -24,7 +24,7 @@ def build_pipeline():
     pipeline.add_op(load_from_hub_op)
 
     caption_images_op = ComponentOp(  
-        component_spec_path="components/captioning_component/fondant_component.yaml",  
+        component_dir="components/captioning_component",  
         arguments={  
             "model_id": "Salesforce/blip-image-captioning-base",  
             "batch_size": 2,  

examples/pipelines/controlnet-interior-design/pipeline.py

@@ -15,7 +15,7 @@
 
 # Define component ops
 generate_prompts_op = ComponentOp(
-    component_spec_path="components/generate_prompts/fondant_component.yaml",
+    component_dir="components/generate_prompts",
     arguments={"n_rows_to_load": None},
 )
 laion_retrieval_op = ComponentOp.from_registry(
@@ -59,9 +59,8 @@
     node_pool_name="model-inference-pool",
 )
 
-write_to_hub_controlnet = ComponentOp.from_registry(
-    name="write_to_hf_hub",
-    component_spec_path="components/write_to_hub_controlnet/fondant_component.yaml",
+write_to_hub_controlnet = ComponentOp(
+    component_dir="components/write_to_hub_controlnet",
     arguments={
         "username": "test-user",
         "dataset_name": "segmentation_kfp",

examples/pipelines/datacomp/pipeline.py

@@ -34,9 +34,8 @@
     "clip_l14_similarity_score": "image_text_clip_l14_similarity_score",
 }
 
-load_from_hub_op = ComponentOp.from_registry(
-    name="load_from_hf_hub",
-    component_spec_path="components/load_from_hf_hub/fondant_component.yaml",
+load_from_hub_op = ComponentOp(
+    component_dir="components/load_from_hf_hub",
     arguments={
         "dataset_name": "nielsr/datacomp-small-with-embeddings",
         "column_name_mapping": load_component_column_mapping,
@@ -48,17 +47,16 @@
     arguments={"min_image_dim": 200, "max_aspect_ratio": 3},
 )
 filter_complexity_op = ComponentOp(
-    component_spec_path="components/filter_text_complexity/fondant_component.yaml",
+    component_dir="components/filter_text_complexity",
     arguments={
         "spacy_pipeline": "en_core_web_sm",
         "batch_size": 1000,
         "min_complexity": 1,
         "min_num_actions": 1,
     },
 )
-cluster_image_embeddings_op = ComponentOp.from_registry(
-    name="cluster_image_embeddings",
-    component_spec_path="components/cluster_image_embeddings/fondant_component.yaml",
+cluster_image_embeddings_op = ComponentOp(
+    component_dir="components/cluster_image_embeddings",
     arguments={
         "sample_ratio": 0.3,
         "num_clusters": 3,

examples/pipelines/finetune_stable_diffusion/pipeline.py

@@ -19,9 +19,8 @@
     value: key for key, value in load_component_column_mapping.items()
 }
 # Define component ops
-load_from_hub_op = ComponentOp.from_registry(
-    name="load_from_hf_hub",
-    component_spec_path="components/load_from_hf_hub/fondant_component.yaml",
+load_from_hub_op = ComponentOp(
+    component_dir="components/load_from_hf_hub",
     arguments={
         "dataset_name": "logo-wizard/modern-logo-dataset",
         "column_name_mapping": load_component_column_mapping,
@@ -72,9 +71,8 @@
     node_pool_name="model-inference-pool",
 )
 
-write_to_hub = ComponentOp.from_registry(
-    name="write_to_hf_hub",
-    component_spec_path="components/write_to_hf_hub/fondant_component.yaml",
+write_to_hub = ComponentOp(
+    component_dir="components/write_to_hf_hub",
     arguments={
         "username": "test-user",
         "dataset_name": "stable_diffusion_processed",

examples/pipelines/starcoder/pipeline.py

@@ -36,9 +36,8 @@
 )
 
 # define ops
-load_from_hub_op = ComponentOp.from_registry(
-    name="load_from_hub",
-    component_spec_path="components/load_from_hub/fondant_component.yaml",
+load_from_hub_op = ComponentOp(
+    component_dir="components/load_from_hub",
     arguments={
         "dataset_name": "ml6team/the-stack-smol-python",
         "column_name_mapping": load_component_column_mapping,

mkdocs.yml

@@ -30,13 +30,18 @@ nav:
   - Home: index.md
   - Getting Started: getting_started.md
   - Building a pipeline: pipeline.md
-  - Creating custom components: custom_component.md
-  - Read / write components: generic_component.md
-  - Component spec: component_spec.md
+  - Components:
+    - Components: components.md
+    - Creating custom components: custom_component.md
+    - Read / write components: generic_component.md
+    - Component spec: component_spec.md
   - Data explorer: data_explorer.md
   - Infrastructure: infrastructure.md
   - Manifest: manifest.md
 
+plugins:
+  - mkdocstrings
+
 markdown_extensions:
   - pymdownx.snippets:
       check_paths: true

pyproject.toml

@@ -68,6 +68,7 @@ coveralls = "^3.3.1"
 
 [tool.poetry.group.docs.dependencies]
 mkdocs-material = "^9.1.8"
+mkdocstrings = { version = "^0.20", extras = ["python"]}
 
 [tool.poetry.scripts]
 fondant = "fondant.cli:entrypoint"

scripts/pre-build.sh

@@ -9,5 +9,5 @@ root_path=$(dirname "$scripts_path")
 
 pushd "$root_path"
 rm -rf src/fondant/components
-cp -r components src/fondant/
+find components/ -type f | grep -i yaml$ | xargs -i cp --parents {} src/fondant/
 popd

src/fondant/compiler.py

@@ -170,10 +170,11 @@ def _generate_spec(self, pipeline: Pipeline, extra_volumes: list) -> dict:
                 "volumes": volumes,
             }
 
-            if component_op.local_component:
-                services[safe_component_name][
-                    "build"
-                ] = f"./{Path(component_op.component_spec_path).parent}"
+            if component_op.dockerfile_path is not None:
+                logger.info(
+                    f"Found Dockerfile for {component_name}, adding build step.",
+                )
+                services[safe_component_name]["build"] = str(component_op.component_dir)
             else:
                 services[safe_component_name][
                     "image"