Merge pull request #94 from monarch-initiative/add_bootstrap

Add Bootstrap function to app; fix some parsing bugs

pyproject.toml

@@ -75,8 +75,8 @@ sphinx-click = {version = ">=4.3.0"}
 myst-parser = {version = ">=0.18.1"}
 
 [tool.poetry.scripts]
-curategpt = "curate_gpt.cli:main"
-gocampr = "curate_gpt.adhoc.gocam_predictor:main"
+curategpt = "curategpt.cli:main"
+gocampr = "curategpt.adhoc.gocam_predictor:main"
 
 [tool.poetry.extras]
 docs = [

src/curategpt/app/app.py

@@ -11,6 +11,7 @@
 
 from curategpt import BasicExtractor
 from curategpt.agents import MappingAgent
+from curategpt.agents.bootstrap_agent import BootstrapAgent, KnowledgeBaseSpecification
 from curategpt.agents.chat_agent import ChatAgent, ChatResponse
 from curategpt.agents.dase_agent import DatabaseAugmentedStructuredExtraction
 from curategpt.agents.dragon_agent import DragonAgent
@@ -37,6 +38,7 @@
 SEARCH = "Search"
 CLUSTER_SEARCH = "Cluster Search"
 MATCH = "Match"
+BOOTSTRAP = "Bootstrap"
 CURATE = "Curate"
 ADD_TO_CART = "Add to Cart"
 # EXTRACT = "Extract"
@@ -77,6 +79,7 @@
     EXTRACT,
     CITESEEK,
     MATCH,
+    BOOTSTRAP,
     CART,
     ABOUT,
     HELP,
@@ -664,6 +667,51 @@ def _flat(obj: dict, limit=40) -> dict:
                     st.subheader(f"Reference {ref}", anchor=f"ref-{ref}")
                     st.code(text, language="yaml")
 
+elif option == BOOTSTRAP:
+    page_state = state.get_page_state(BOOTSTRAP)
+    st.subheader("Generate a schema and data for a new knowledge base.")
+
+    extractor = BasicExtractor()
+    extractor.model_name = model_name
+    bootstrap_agent = BootstrapAgent(extractor=extractor)
+
+    kb_name = st.text_input("KB Name", help="Name of the knowledge base, without spaces (e.g. 'ice_cream_kb')")
+    description = st.text_input(
+        "Description",
+        help="Description of the knowledge base (e.g. 'A knowledge base for ice cream')",
+    )
+    attributes = st.text_input(
+        "Attributes",
+        help="Attributes of the knowledge base (e.g. 'flavor, viscosity, color')",
+    )
+    main_class = st.text_input(
+        "Main Class",
+        help="Main class of the knowledge base, without spaces (e.g. 'IceCreamType')",
+    )
+    generate_data = st.checkbox(
+        "Generate data",
+        help="""
+        If checked, after generating the schema, generate example data.
+        """,
+    )
+    if st.button("Make Schema"):
+        st.write(f"Generating schema for *{kb_name}*")
+        config_dict = {
+            "kb_name": kb_name,
+            "description": description,
+            "attributes": attributes,
+            "main_class": main_class,
+        }
+        config = KnowledgeBaseSpecification(**config_dict)
+        ao = bootstrap_agent.bootstrap_schema(config)
+        schema_dict = ao.model_dump()
+        st.write(schema_dict)
+
+        if generate_data:
+            st.write(f"Generating data for *{kb_name}*")
+            data = bootstrap_agent.bootstrap_data(schema=schema_dict)
+            st.code(data, language="yaml")
+
 elif option == CART:
     page_state = state.get_page_state(CART)
     st.subheader("Your items")

src/curategpt/cli.py

@@ -829,7 +829,32 @@ def extract_from_pubmed(
 
 @main.group()
 def bootstrap():
-    "Bootstrap schema or data."
+    """Bootstrap schema or data.
+
+    Starting with a general description or a LinkML schema,
+    generate an initial version of a knowledge base.
+
+    The config should be a yaml file with the following fields:
+    kb_name: str
+    description: str
+    attributes: str
+    main_class: str
+
+    For example, this is a valid config:
+    kb_name: lumber_kb
+    description: A knowledge base for lumber
+    attributes: source_tree
+    main_class: Lumber_Type
+
+    Examples:
+
+    curategpt bootstrap schema -C config.yaml
+    (This will generate a LinkML schema, based on the provided config.)
+
+    curategpt bootstrap data -s schema.yaml
+    (This will generate data based on the provided schema.
+    The output of the previous command can be used as input for this command.)
+    """
 
 
 @bootstrap.command(name="schema")
@@ -841,7 +866,7 @@ def bootstrap():
     help="path to yaml config",
 )
 def bootstrap_schema(config, model):
-    """Bootstrap a knowledge base."""
+    """Bootstrap a knowledge base with LinkML schema."""
     extractor = BasicExtractor()
     if model:
         extractor.model_name = model
@@ -865,7 +890,7 @@ def bootstrap_schema(config, model):
     help="path to yaml linkml schema",
 )
 def bootstrap_data(config, schema, model):
-    """Bootstrap a knowledge base."""
+    """Bootstrap a knowledge base with initial data."""
     extractor = BasicExtractor()
     if model:
         extractor.model_name = model

src/curategpt/conf/prompts/bootstrap-schema.j2

@@ -36,7 +36,7 @@ classes:
 
 The goal is to have a tree like model with {{  main_class }} as the root of the tree,
 and then lists of rich objects as the children of the root. Each rich object may need
-its own class to describe it
+its own class to describe it. Attributes of the main class should include {{ attributes }}.
 
 ## Examples
 
@@ -151,6 +151,9 @@ classes:
 
 ## General Tips
 
+- Do not include any markdown formatting such as code fences, except as specified above
+- Do not include any commentary preceding the YAML
+- Do not include any "`" characters within the YAML
 - don't include base types like string etc under `types`, they are imported already
 - give {{ main_class }} a mixture of simple string metadata and lists of rich objects for associated data that might require provenance or other metadata.
 - make sure the YAML is complete. Even though I provide placeholder "..." in the examples, the YAML you provide must parse