#!/usr/bin/env python3
"""
Ad-hoc test of different Qdrant vectors for card DSL parsing.
Tests:
1. 'components' vector - component identity search
2. 'inputs' vector - content/parameter search
3. 'relationships' vector - structure/DSL search
Goal: Understand what each vector returns and how to simplify DSL parsing
by relying more on Qdrant results.
"""
import os
import sys
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from config.qdrant_client import get_qdrant_client
def main():
print("=" * 70)
print("AD-HOC QDRANT VECTOR SEARCH TESTS")
print("=" * 70)
client = get_qdrant_client()
collection = "mcp_gchat_cards_v7"
# Check collection info
info = client.get_collection(collection)
print(f"\nCollection: {collection}")
print(f"Points: {info.points_count}")
print(f"Vectors: {list(info.config.params.vectors.keys())}")
# Test queries
test_queries = [
# DSL symbol queries
"§ Section with decorated text",
"δ DecoratedText with icon",
"Ƀ ButtonList containing buttons",
"ᵬ Button with URL",
# Natural language
"card with title subtitle and buttons",
"section header with icon and text",
# DSL patterns
"§[δ, Ƀ[ᵬ×2]]",
"Ƀ[ᵬ×2]",
]
print("\n" + "=" * 70)
print("1. COMPONENTS VECTOR SEARCH (ColBERT - component identity)")
print("=" * 70)
try:
from fastembed import LateInteractionTextEmbedding
colbert = LateInteractionTextEmbedding(model_name="colbert-ir/colbertv2.0")
for query in test_queries[:4]:
print(f"\n🔍 Query: '{query}'")
vectors = list(colbert.query_embed(query))[0]
vectors_list = [v.tolist() for v in vectors]
from qdrant_client import models
results = client.query_points(
collection_name=collection,
query=vectors_list,
using="components",
query_filter=models.Filter(
must=[
models.FieldCondition(
key="type", match=models.MatchValue(value="class")
)
]
),
limit=3,
with_payload=True,
)
for p in results.points:
print(
f" • {p.payload.get('name', '?')} (score: {p.score:.3f}) - {p.payload.get('type')}"
)
if p.payload.get("relationship_text"):
print(f" rel: {p.payload.get('relationship_text')[:60]}...")
except Exception as e:
print(f" ❌ ColBERT search failed: {e}")
print("\n" + "=" * 70)
print("2. INPUTS VECTOR SEARCH (ColBERT - content/parameters)")
print("=" * 70)
try:
for query in test_queries[:4]:
print(f"\n🔍 Query: '{query}'")
vectors = list(colbert.query_embed(query))[0]
vectors_list = [v.tolist() for v in vectors]
results = client.query_points(
collection_name=collection,
query=vectors_list,
using="inputs",
query_filter=models.Filter(
must=[
models.FieldCondition(
key="type",
match=models.MatchValue(value="instance_pattern"),
)
]
),
limit=3,
with_payload=True,
)
for p in results.points:
name = p.payload.get("name", "?")
feedback = p.payload.get("content_feedback", p.payload.get("feedback"))
struct = p.payload.get("structure_description", "")[:50]
print(f" • {name} (score: {p.score:.3f}) feedback={feedback}")
if struct:
print(f" struct: {struct}...")
except Exception as e:
print(f" ❌ Inputs search failed: {e}")
print("\n" + "=" * 70)
print("3. RELATIONSHIPS VECTOR SEARCH (MiniLM - structure)")
print("=" * 70)
try:
from fastembed import TextEmbedding
minilm = TextEmbedding(model_name="sentence-transformers/all-MiniLM-L6-v2")
for query in test_queries:
print(f"\n🔍 Query: '{query}'")
vector = list(minilm.embed([query]))[0].tolist()
results = client.query_points(
collection_name=collection,
query=vector,
using="relationships",
limit=3,
with_payload=True,
)
for p in results.points:
name = p.payload.get("name", "?")
ptype = p.payload.get("type", "?")
rel_text = p.payload.get("relationship_text", "")[:60]
print(f" • {name} (score: {p.score:.3f}) type={ptype}")
if rel_text:
print(f" rel_text: {rel_text}...")
except Exception as e:
print(f" ❌ Relationships search failed: {e}")
print("\n" + "=" * 70)
print("4. INSTANCE PATTERNS WITH POSITIVE FEEDBACK")
print("=" * 70)
try:
# Just scroll and show some positive feedback patterns
results, _ = client.scroll(
collection_name=collection,
scroll_filter=models.Filter(
must=[
models.FieldCondition(
key="type", match=models.MatchValue(value="instance_pattern")
),
models.FieldCondition(
key="content_feedback",
match=models.MatchValue(value="positive"),
),
]
),
limit=5,
with_payload=True,
)
print(f"\n Found {len(results)} positive content patterns:")
for p in results:
name = p.payload.get("name", "?")
struct = p.payload.get("structure_description", "")[:80]
comp_paths = p.payload.get("component_paths", [])
print(f"\n • {name}")
print(f" structure: {struct}")
print(f" paths: {comp_paths[:5]}")
except Exception as e:
print(f" ❌ Pattern scroll failed: {e}")
print("\n" + "=" * 70)
print("5. DSL-ANNOTATED PATTERNS (checking for symbol-enriched data)")
print("=" * 70)
try:
# Search for patterns that have DSL symbols in relationship_text
results, _ = client.scroll(
collection_name=collection,
scroll_filter=models.Filter(
must=[
models.FieldCondition(
key="relationship_text",
match=models.MatchText(text="§"), # Section symbol
)
]
),
limit=5,
with_payload=True,
)
print(f"\n Found {len(results)} patterns with § symbol:")
for p in results:
name = p.payload.get("name", "?")
rel_text = p.payload.get("relationship_text", "")[:100]
print(f"\n • {name}")
print(f" rel_text: {rel_text}")
# Search for δ (DecoratedText)
results, _ = client.scroll(
collection_name=collection,
scroll_filter=models.Filter(
must=[
models.FieldCondition(
key="relationship_text",
match=models.MatchText(text="δ"),
)
]
),
limit=3,
with_payload=True,
)
print(f"\n Found {len(results)} patterns with δ symbol:")
for p in results:
print(
f" • {p.payload.get('name', '?')}: {p.payload.get('relationship_text', '')[:80]}"
)
except Exception as e:
print(f" ❌ DSL pattern search failed: {e}")
print("\n" + "=" * 70)
print("TEST COMPLETE")
print("=" * 70)
if __name__ == "__main__":
main()
