hweller1

## hybrid_search_example.py

import os

import pymongo

os.environ["OPENAI_API_KEY"] = '<openai API key>'
import openai


### SETUP

## parent_document_retriever.py
import os
import pymongo
import time

import openai

embeddings = openai.Embedding.create(
  input="What is a transformer?",
  model="text-embedding-ada-002"
)

## generate_exact_result_set.py
import numpy as np
from sentence_transformers import SentenceTransformer

queries = [f"What is {x}" for x in names]
encoded_queries = {}
model = SentenceTransformer('sentence-transformers/facebook-dpr-question_encoder-single-nq-base')
results = {}
top_result = []
for i, query in enumerate(queries):
    print(f"Calculating similarity for query {i}")

## companies.py
names = ["Qualcomm",
"Hewlett Packard Enterprise",
"British American Tobacco",
"Visa",
"China Pacific Insurance",
"MetLife",
"AstraZeneca",
"Altria Group",
"SAP",
"Costco Wholesale",

## rsf_via_union_with.py

import pymongo
import time
from sentence_transformers import SentenceTransformer

from companies import names # List of company names from another python file


### DESCRIPTION
"""

## rrf_via_unionWith.py
import pymongo
import time
from sentence_transformers import SentenceTransformer


from companies import names # list of company names in a separate python file

### DESCRIPTION
"""
Search against the Sphere dataset using vector search results fused with full text search results via reciprocal rank fusion.

## compute_overlap.py
def compute_overlap(exact_result_set: List, approx_result_set: List) -> float:
	# each result set is a list of urls, order not considered
	return len(exact_result_set.intersection(set(approx_result_set)) / len(set(exact_result_set))

## filter_experiments.py
$search": {
	            "index":'1M_sphere_index',
	            "knnBeta": {
	            "path": "vector",
	            "vector": embedding.tolist(),
	            "k": k * multiplier,
	            # "filter":{
          		# 	"equals":{
	            # 	"path":"low_card",
		        #     "value":1

## sample_index_definition.py
{
  "type": "vectorSearch,
  "fields": [{
    "path": "plot_embedding",
    "dimensions": 1536,
    "similarity": "cosine",
    "type": "vector"
  }]
}

## sample_index2.py
{
  "type": "vectorSearch,
  "fields": [{
    "path": "plot_embedding_hf",
    "dimensions": 384,
    "similarity": "dotProduct",
    "type": "vector"
  }]
}

	import os

	import pymongo

	os.environ["OPENAI_API_KEY"] = '<openai API key>'
	import openai


	### SETUP
	import os
	import pymongo
	import time

	import openai

	embeddings = openai.Embedding.create(
	input="What is a transformer?",
	model="text-embedding-ada-002"
	)
	import numpy as np
	from sentence_transformers import SentenceTransformer

	queries = [f"What is {x}" for x in names]
	encoded_queries = {}
	model = SentenceTransformer('sentence-transformers/facebook-dpr-question_encoder-single-nq-base')
	results = {}
	top_result = []
	for i, query in enumerate(queries):
	print(f"Calculating similarity for query {i}")
	names = ["Qualcomm",
	"Hewlett Packard Enterprise",
	"British American Tobacco",
	"Visa",
	"China Pacific Insurance",
	"MetLife",
	"AstraZeneca",
	"Altria Group",
	"SAP",
	"Costco Wholesale",
	def compute_overlap(exact_result_set: List, approx_result_set: List) -> float:
	# each result set is a list of urls, order not considered
	return len(exact_result_set.intersection(set(approx_result_set)) / len(set(exact_result_set))
	$search": {
	"index":'1M_sphere_index',
	"knnBeta": {
	"path": "vector",
	"vector": embedding.tolist(),
	"k": k * multiplier,
	# "filter":{
	# "equals":{
	# "path":"low_card",
	# "value":1
	{
	"type": "vectorSearch,
	"fields": [{
	"path": "plot_embedding",
	"dimensions": 1536,
	"similarity": "cosine",
	"type": "vector"
	}]
	}
	{
	"type": "vectorSearch,
	"fields": [{
	"path": "plot_embedding_hf",
	"dimensions": 384,
	"similarity": "dotProduct",
	"type": "vector"
	}]
	}