media_assistant / test_default.py
uijnbh's picture
Rebuilt S3 interface with inventory
6ab18b3
Raw History Blame Contribute Delete
2.71 kB
from dotenv import load_dotenv
from pinecone import Pinecone
from transformers import AutoTokenizer, AutoModel
import torch
import pandas as pd
import pandas.testing as pdt
import boto3
import os
from io import StringIO
from gradio_client import Client
load_dotenv()
MODEL_NAME = "sentence-transformers/all-mpnet-base-v2"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModel.from_pretrained(MODEL_NAME)
pc = Pinecone(api_key=os.environ["PINECONE_KEY"])
index = pc.Index("product-recommender")
def compare_s3_to_setupfiles(file_name: str) -> bool:
expected_table = pd.read_csv("setupfiles/test_" + file_name)
actual_table = pd.read_csv(
StringIO(
boto3.client("s3")
.get_object(Bucket="charityshop-media-assistant", Key=file_name)["Body"]
.read()
.decode("utf-8")
)
)
actual_test_table = actual_table[
actual_table.store_name == "TestStore"
].reset_index(drop=True)
return pdt.assert_frame_equal(expected_table, actual_test_table)
def text_to_embedding(text):
inputs = tokenizer(text, padding=True, truncation=True, return_tensors="pt")
with torch.no_grad():
outputs = model(**inputs)
embeddings = outputs.last_hidden_state.mean(dim=1).numpy()
return embeddings.flatten().tolist()
def find_similar_media(query_text, top_k=1):
query_embedding = text_to_embedding(query_text)
matches = index.query(
vector=query_embedding,
top_k=top_k,
include_metadata=True,
filter={"vector_type": "test"},
)["matches"]
df = pd.DataFrame(
{"barcode": [m["id"] for m in matches], "score": [m["score"] for m in matches]}
)
for col in ["artist", "title", "media_type", "description", "vector_type"]:
df[col] = [m["metadata"][col] for m in matches]
return df
def test_InventoryDatabaseMatchesSampleRecords():
compare_s3_to_setupfiles("inventory.csv")
def test_ChatHistoryDatabaseMatchesSampleRecords():
compare_s3_to_setupfiles("chat_history.csv")
def test_VectorDatabaseMatchesSampleRecords():
expected_table = pd.read_csv(
"setupfiles/test_vector_database_query.csv", dtype={"barcode": str}
)
actual_test_table = find_similar_media("denver", top_k=5).drop(columns=["score"])
pdt.assert_frame_equal(expected_table, actual_test_table)
def test_CanQueryLLM():
current_site = "media_assistant"
client = Client("ml-usecases/" + current_site, hf_token=os.environ["HF_TOKEN"])
try:
result = client.predict(
user_query="Hello!!", api_name="/combined_recommendation"
)
return len(str(result)) > 0
except:
return False