155 lines
5.8 KiB
Python
155 lines
5.8 KiB
Python
import chromadb
|
|
from chromadb.config import Settings
|
|
from openai import OpenAI
|
|
import os
|
|
|
|
|
|
class FinancialSituationMemory:
|
|
def __init__(self, name, config):
|
|
self.provider = config.get("llm_provider", "openai").lower()
|
|
|
|
# Determine embedding model based on provider
|
|
if self.provider == "ollama":
|
|
self.embedding = "nomic-embed-text"
|
|
self.use_ollama_embeddings = True
|
|
elif config.get("backend_url") == "http://localhost:11434/v1":
|
|
self.embedding = "nomic-embed-text"
|
|
self.use_ollama_embeddings = True
|
|
else:
|
|
self.embedding = "text-embedding-3-small"
|
|
self.use_ollama_embeddings = False
|
|
|
|
# Only create OpenAI client if we're using OpenAI embeddings
|
|
if not self.use_ollama_embeddings:
|
|
api_key = os.getenv("OPENAI_API_KEY")
|
|
if api_key:
|
|
self.client = OpenAI(base_url=config.get("backend_url"), api_key=api_key)
|
|
else:
|
|
self.client = OpenAI(base_url=config.get("backend_url"))
|
|
else:
|
|
self.client = None
|
|
|
|
self.chroma_client = chromadb.Client(Settings(allow_reset=True))
|
|
self.situation_collection = self.chroma_client.create_collection(name=name)
|
|
|
|
def get_embedding(self, text):
|
|
"""Get embedding for text - provider agnostic"""
|
|
|
|
if self.use_ollama_embeddings:
|
|
# For Ollama, use chromadb's built-in embedding function
|
|
# or return None to let chromadb handle it
|
|
# ChromaDB will use its default embedding function
|
|
return None
|
|
else:
|
|
# Use OpenAI embeddings
|
|
response = self.client.embeddings.create(
|
|
model=self.embedding, input=text
|
|
)
|
|
return response.data[0].embedding
|
|
|
|
def add_situations(self, situations_and_advice):
|
|
"""Add financial situations and their corresponding advice. Parameter is a list of tuples (situation, rec)"""
|
|
|
|
situations = []
|
|
advice = []
|
|
ids = []
|
|
embeddings = []
|
|
|
|
offset = self.situation_collection.count()
|
|
|
|
for i, (situation, recommendation) in enumerate(situations_and_advice):
|
|
situations.append(situation)
|
|
advice.append(recommendation)
|
|
ids.append(str(offset + i))
|
|
embedding = self.get_embedding(situation)
|
|
if embedding is not None:
|
|
embeddings.append(embedding)
|
|
|
|
# Add to collection - chromadb will use default embeddings if none provided
|
|
add_params = {
|
|
"documents": situations,
|
|
"metadatas": [{"recommendation": rec} for rec in advice],
|
|
"ids": ids,
|
|
}
|
|
|
|
if embeddings: # Only add embeddings if we have them
|
|
add_params["embeddings"] = embeddings
|
|
|
|
self.situation_collection.add(**add_params)
|
|
|
|
def get_memories(self, current_situation, n_matches=1):
|
|
"""Find matching recommendations using embeddings"""
|
|
query_embedding = self.get_embedding(current_situation)
|
|
|
|
# Build query parameters
|
|
query_params = {
|
|
"n_results": n_matches,
|
|
"include": ["metadatas", "documents", "distances"],
|
|
}
|
|
|
|
if query_embedding is not None:
|
|
query_params["query_embeddings"] = [query_embedding]
|
|
else:
|
|
# Use text-based search if no embeddings
|
|
query_params["query_texts"] = [current_situation]
|
|
|
|
results = self.situation_collection.query(**query_params)
|
|
|
|
matched_results = []
|
|
for i in range(len(results["documents"][0])):
|
|
matched_results.append(
|
|
{
|
|
"matched_situation": results["documents"][0][i],
|
|
"recommendation": results["metadatas"][0][i]["recommendation"],
|
|
"similarity_score": 1 - results["distances"][0][i],
|
|
}
|
|
)
|
|
|
|
return matched_results
|
|
|
|
|
|
if __name__ == "__main__":
|
|
# Example usage
|
|
matcher = FinancialSituationMemory()
|
|
|
|
# Example data
|
|
example_data = [
|
|
(
|
|
"High inflation rate with rising interest rates and declining consumer spending",
|
|
"Consider defensive sectors like consumer staples and utilities. Review fixed-income portfolio duration.",
|
|
),
|
|
(
|
|
"Tech sector showing high volatility with increasing institutional selling pressure",
|
|
"Reduce exposure to high-growth tech stocks. Look for value opportunities in established tech companies with strong cash flows.",
|
|
),
|
|
(
|
|
"Strong dollar affecting emerging markets with increasing forex volatility",
|
|
"Hedge currency exposure in international positions. Consider reducing allocation to emerging market debt.",
|
|
),
|
|
(
|
|
"Market showing signs of sector rotation with rising yields",
|
|
"Rebalance portfolio to maintain target allocations. Consider increasing exposure to sectors benefiting from higher rates.",
|
|
),
|
|
]
|
|
|
|
# Add the example situations and recommendations
|
|
matcher.add_situations(example_data)
|
|
|
|
# Example query
|
|
current_situation = """
|
|
Market showing increased volatility in tech sector, with institutional investors
|
|
reducing positions and rising interest rates affecting growth stock valuations
|
|
"""
|
|
|
|
try:
|
|
recommendations = matcher.get_memories(current_situation, n_matches=2)
|
|
|
|
for i, rec in enumerate(recommendations, 1):
|
|
print(f"\nMatch {i}:")
|
|
print(f"Similarity Score: {rec['similarity_score']:.2f}")
|
|
print(f"Matched Situation: {rec['matched_situation']}")
|
|
print(f"Recommendation: {rec['recommendation']}")
|
|
|
|
except Exception as e:
|
|
print(f"Error during recommendation: {str(e)}")
|