QA with Embeding

This is an attempt to build simple Question Answer Systems for data about real estate property.

Drawn heavily from: https://github.com/openai/openai-cookbook/blob/main/examples/Question_answering_using_embeddings.ipynb

It uses the Flask web framework.

Setup the docker compose (After cloning this repo and changing current direcrtory into the repo directory)

1. Make a copy of the example environment variables file:

   $ cp .env.example .env

2. Add your API key to the newly created .env file.

3. Build the docker image with dependencies:

docker build . -t openai

4. Start docker compose with openia and clickhouse containers:

U_ID=$UID docker-compose up --remove-orphans

Observe the succesful start and open another terminal. To make sure containers are up execute the docker ps in the another terminal:

$ docker ps
CONTAINER ID   IMAGE                          COMMAND            CREATED       STATUS         PORTS                                                                                            NAMES
21254e03dc9c   openai                         "bash"             2 hours ago   Up 2 minutes   0.0.0.0:5000->5000/tcp, :::5000->5000/tcp                                                        openai
c20a49fa2f36   clickhouse/clickhouse-server   "/entrypoint.sh"   3 hours ago   Up 2 minutes   0.0.0.0:8123->8123/tcp, :::8123->8123/tcp, 0.0.0.0:9000->9000/tcp, :::9000->9000/tcp, 9009/tcp   clickhouse

Create clickhouse schema and load the data

The file addess_embeddings_descriptions.csv.gz contains sample data with embedings vectors precomputed for an address field. The sample data must be loaded into clickhouse db. The docker compose uses persistent volume to store clickhouse data and data survives the restarts of docker compose.

Create the schema in clickhouse

Make sure clickhouse-client is installed https://clickhouse.com/docs/en/integrations/sql-clients/cli. Connect to clickhouse:

clickhouse-client -u lupi --password lupi -d openai

If there is no errors this is what should be on the screen:

$ clickhouse-client -u lupi --password lupi -d openai 
ClickHouse client version 23.7.4.5 (official build).
Connecting to database openai at localhost:9000 as user lupi.
Connected to ClickHouse server version 23.7.4 revision 54465.

clickhouse :) 

Paste the following to create a table:

CREATE TABLE qa_properties
(
	`address` String,
    `description` String,
    `address_embeddings` Array(Float32)
)
ENGINE = MergeTree
ORDER BY (`address_embeddings`,`address`)

Verify the table is created:

clickhouse :) show tables

SHOW TABLES

Query id: 9bfd7044-c774-486a-81c4-ce470081080c

┌─name──────────┐
│ qa_properties │
└───────────────┘

1 row in set. Elapsed: 0.006 sec. 

clickhouse :) desc qa_properties

DESCRIBE TABLE qa_properties

Query id: 6db0efd8-9760-4fd4-a562-5b92ef46ad3d

┌─name───────────────┬─type───────────┬─default_type─┬─default_expression─┬─comment─┬─codec_expression─┬─ttl_expression─┐
│ address            │ String         │              │                    │         │                  │                │
│ description        │ String         │              │                    │         │                  │                │
│ address_embeddings │ Array(Float32) │              │                    │         │                  │                │
└────────────────────┴────────────────┴──────────────┴────────────────────┴─────────┴──────────────────┴────────────────┘

3 rows in set. Elapsed: 0.004 sec. 

Load the data to clickhouse table

The loading is performed by connecting to openai container and running python commands.

Connect to running instance of openai container:

docker exec -it openai bash

Start python3 interpreter. In the python interpreter run:

>>> import app
>>> app.save_table_to_db()
              address                                        description                                 address_embeddings
0       21 Pioneer Rd  Price: $1,243,680\n2 bd 3 ba\n2,482 sqft\n21 P...  [-0.0036932125, 0.004790107, 0.0033191952, 0.0...
1  167 Parkerville Rd  Price: $575,000\n3 bd 2 ba\n1,644 sqft\n167 Pa...  [-0.0033807599, 0.017201519, -0.0011668914, -0...
                address                                        description                                 address_embeddings
0         21 Pioneer Rd  Price: $1,243,680\n2 bd 3 ba\n2,482 sqft\n21 P...  [-0.0036932125, 0.004790107, 0.0033191952, 0.0...
1    167 Parkerville Rd  Price: $575,000\n3 bd 2 ba\n1,644 sqft\n167 Pa...  [-0.0033807599, 0.017201519, -0.0011668914, -0...
2        3 Wyndemere Dr  price: $2,999,999\n4 bd 7 ba\n7,041 sqft\n3 Wy...  [-0.017820623, -0.0067525995, -0.020082321, -0...
3          14 Joslin Ln  Price: $2,195,000\n4 bd 5 ba\n6,698 sqft\n14 J...  [-0.015576344, 0.027265169, -0.008694384, 0.00...
4    167 Parkerville Rd  Price: $575,000\n3 bd 2 ba\n1,644 sqft\n167 Pa...  [-0.0033807599, 0.017201519, -0.0011668914, -0...
5          44 Oregon Rd  Price: $999,900\n4 bd 3 ba\n2,810 sqft\n44 Ore...  [-0.008821211, 0.0020653352, -0.021821918, 0.0...
6           37 James Rd  Price: $2,100,000\n14.81 Acres\n37 James Rd, A...  [-0.00501842, 0.03608214, -0.02601209, 0.00843...
7        16 Bridle Path  Price: $1,297,000\n6 bd 6 ba\n4,844 sqft\n16 B...  [-0.021653378, 0.0073725246, -0.024657, -0.009...
8  22 Wachusett View Dr  Price: $1,250,000\n5 bd 4 ba\n3,340 sqft\n22 W...  [-0.008583948, 0.021219622, 0.0023700008, 0.00...
9          12 Brooke St  Price: $1,339,520\n4 bd 4 ba\n4,019 sqft\n12 B...  [-0.013631155, -0.009619288, 0.0025738983, -0....
>>> 
root@openai:/app# 

Observe the output of dataframe going to database.

Connect to clickhouse and select one row to confirm the data is in the tables:

$ clickhouse-client -u lupi --password lupi -d openai 
ClickHouse client version 23.7.4.5 (official build).
Connecting to database openai at localhost:9000 as user lupi.
Connected to ClickHouse server version 23.7.4 revision 54465.

clickhouse :) select address_embeddings from qa_properties limit 1

SELECT address_embeddings
FROM qa_properties
LIMIT 1

Query id: aa9d23f9-b936-46fa-87a4-e9413b1c8c45

┌─address_embeddings─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ [-0.021653378,0.0073725246,-0.024657,-0.009843686,0.0039047075,0.004723877,-0.0085398415,-0.013161322,0.0035838662,-0.01668375,-0.007427136,-0.007488574,-0.010362493,0.013270545,-0.00068562775,0.025407905,0.01402145,0.009966562,0.0062427535,-0.010212312,-0.042815253,0.013147669,0.019414315,-0.016042069,0.003955906,...<SKIPPED A LOT>...,0.026704922,0.023551121,0.012485507,-0.011400107,-0.0027612837,-0.020711333,0.025694614,-0.010963217,-0.01150933,-0.0056420295,0.0011707296,-0.014089714,0.025407905,-0.028943986,0.017161598,0.0010598004,-0.017366393,0.0070994683,-0.010546806,0.0012697126,-0.0059935898,0.0017398817,0.046774574,-0.011522983,0.01044441,0.0059082597,0.017612142,-0.02785176,0.0074476153,0.0062871254,0.025749225,-0.010826689,0.0041709375,-0.0030531127,-0.00040403826,⋯│
└────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

1 row in set. Elapsed: 0.008 sec. 

clickhouse :)

Making queries, Working with web interface.

You should now be able to access the app at http://localhost:5000

Using postman

Send the following JSON using POST to localhost:5000/property_data

{"question": "what is the price of 22 Wachusett View Dr"}