Add docs, add pdm deps

adala/skills/collection/rag.py

@@ -10,6 +10,34 @@
 class RAGSkill(TransformSkill):
     """
     Skill for RAG (Retrieval-Augmented Generation) models.
+
+    Attributes:
+        input_template: Template for the input. It wraps the input with the template to create a query.
+        rag_input_template: Template for RAG input. It wraps each retrieved item with the template, and then concatenates them with two newlines.
+                            Example: "Question: {question}\nContext: {context}" with num_results=2 will result in "Question: <question>\nContext: <context>\n\nQuestion: <question>\nContext: <context>"
+        instructions: Instructions for the generation part of the RAG model.
+        output_template: Template for the output. It wraps the output with the template.
+        num_results: Number of results to retrieve from the memory.
+        memory: Memory to use for retrieval. If None, a VectorDBMemory will be used.
+
+    Examples:
+        >>> from adala.skills import RAGSkill
+        >>> skill = RAGSkill(
+        ...     name="rag",
+        ...     input_template="Question: {question}",
+        ...     rag_input_template="Question: {question}\nContext: {context}",
+        ...     instructions="Answer the question.",
+        ...     output_template="{answer}",
+        ...     num_results=2,
+        ... )
+        >>> skill.apply(
+        ...     input=InternalDataFrame(
+        ...         data=[
+        ...             {"question": "What is the meaning of life?", "context": "Life is a game."},
+        ...             {"question": "What is the meaning of life?", "context": "Life is a game."},
+        ...         ]
+        ...     ))
+
     """
 
     name: str = "rag"
@@ -32,6 +60,15 @@ def apply(
     ) -> InternalDataFrame:
         """
         Apply the skill.
+
+        Args:
+            input: Input data.
+            runtime: Runtime to use for generation.
+
+        Returns:
+            Output data. The output field is named after the output_template.
+                        If no instructions are given, the output field contains concatenated strings from retrieved items.
+                        If instructions are given, the output field contains the generated output.
         """
         input_strings = input.apply(
             lambda r: self.input_template.format(**r), axis=1
@@ -51,13 +88,15 @@ def apply(
         output_field = output_fields[0]
         rag_input = InternalDataFrame({output_field: rag_input_strings})
         if self.instructions:
+            # if instructions are given, use the runtime to generate the output
             output = runtime.batch_to_batch(
                 rag_input,
                 instructions_template=self.instructions,
                 input_template=f"{{{output_field}}}",
                 output_template=self.output_template,
             )
         else:
+            # if no instructions - simply return the rag input
             output = rag_input
             output.index = input.index
 
@@ -71,7 +110,13 @@ def improve(
         runtime: Runtime,
     ):
         """
-        Improve the skill.
+        Improve the skill by storing the feedback match errors in the memory.
+
+        Args:
+            predictions: Predictions made by the skill.
+            train_skill_output: Output field of the skill used for training.
+            feedback: Feedback data. for feedback.match equals False (prediction errors), the input is stored in the memory.
+            runtime: Runtime to use for generation (not used).
         """
 
         error_indices = feedback.match[