paulfitz/mlsql

Infer SQL queries from plain-text questions and table headers.

Requirements:

• install docker
• install curl (or, if you're feeling brave, asql)
• Make sure docker allows at least 3GB of RAM (see Docker>Preferences>Advanced or equivalent) for SQLova, or 5GB for IRNet or ValueNet.

I take pretrained models published along with academic papers, and do whatever it takes to make them testable on fresh data (academic work often omits that, with code tied to a particular benchmark dataset). I spend days tracking down and patching obscure data preprocessing steps so you don't have to.

So far I've packaged three models:

• SQLova. Works on single tables.
• ValueNet. Works on multiple tables, and makes an effort to predict parameters.
• IRNet. Works on multiple tables, but doesn't predict parameters.

In each case, I've mangled the original network somewhat, so if they interest you do follow up with the original sources.

SQLova

This wraps up a published pretrained model for SQLova (https://github.com/naver/sqlova/).

Fetch and start SQLova running as an api server on port 5050:

docker run --name sqlova -d -p 5050:5050 paulfitz/sqlova

Be patient, the image is about 4.2GB. Once it is running, it'll take a few seconds to load models and then you can start asking questions about CSV tables. For example:

curl -F "[email protected]" -F "q=how long is throgs neck" localhost:5050
# {"answer":[1800],"params":["throgs neck"],"sql":"SELECT (length) FROM bridges WHERE bridge = ?"}

This is using the sample bridges.csv included in this repo.

Here are some examples of the answers and sql inferred for plain-text questions about this table:

With the players.csv sample from WikiSQL:

Some questions about iris.csv:

There are plenty of types of questions this model cannot answer (and that aren't covered in the dataset it is trained on, or in the sql it is permitted to generate).

ValueNet

This wraps up a published pretrained model for ValueNet (https://github.com/brunnurs/valuenet).

Fetch and start ValueNet running as an api server on port 5050:

docker run --name valuenet -d -p 5050:5050 paulfitz/valuenet

You can then ask questions of individual csv files as before, or several csv files (just repeat -F "[email protected]") or a simple sqlite db with tables related by foreign keys. In this last case, the model can answer using joins.

curl -F "[email protected]" -F "q=who is the CEO of Omni Cooperative" localhost:5050
# {"answer":[["Dracula"]], "sql":"SELECT T1.name FROM people AS T1 JOIN organizations AS T2 \
#   ON T1.id = T2.ceo_id WHERE T2.company = 'Omni Cooperative'"}
curl -F "[email protected]" -F "q=how many designers are there?" localhost:5050
# {"answer":[[5]],"sql":"SELECT DISTINCT count(DISTINCT T1.designer) FROM bridges AS T1"}
curl -F "[email protected]" -F "[email protected]" -F "q=how many designers are there?" localhost:5050
# same answer
curl -F "[email protected]" -F "[email protected]" -F "q=what is the name of the airport with the highest latitude?" localhost:5050
# {"answer":[["Disraeli Inlet Water Aerodrome"]],
#  "sql":"SELECT T1.name FROM airports AS T1 ORDER BY T1.latitude_deg DESC LIMIT 1"}

I've includes material to convert user tables into the form needed to query them. Don't judge the network by its quality here, go do a deep dive with the original - I've deviated from the original in important respects, including how named entity recognition is done.

I've written up some experiments with ValueNet.

IRNet

This wraps up a published pretrained model for IRNet (https://github.com/microsoft/IRNet). Upstream released a better model after I packaged this, so don't judge the model by playing with it here.

Fetch and start IRNet running as an api server on port 5050:

docker run --name irnet -d -p 5050:5050 -v $PWD/cache:/cache paulfitz/irnet

Be super patient! Especially on the first run, when a few large models need to be downloaded and unpacked.

You can then ask questions of individual csv files as before, or several csv files (just repeat -F "[email protected]") or a simple sqlite db with tables related by foreign keys. In this last case, the model can answer using joins.

curl -F "[email protected]" -F "q=what city is The Firm headquartered in?" localhost:5050
# Answer: SELECT T1.city FROM locations AS T1 JOIN organizations AS T2 WHERE T2.company = 1
curl -F "[email protected]" -F "q=who is the CEO of Omni Cooperative" localhost:5050
# Answer: SELECT T1.name FROM people AS T1 JOIN organizations AS T2 WHERE T2.company = 1
curl -F "[email protected]" -F "q=what company has Dracula as CEO" localhost:5050
# Answer: SELECT T1.company FROM organizations AS T1 JOIN people AS T2 WHERE T2.name = 1

(Note there's no value prediction, so e.g. the where clauses are = 1 rather than something more useful).

Postman users

Curl can be replaced by Postman for those who like that. Here's a working set-up:

Other models

I hope to track research in the area and substitute in models as they become available:

• Spider leaderboard

• WikiSQL leaderboard

• SparC leaderboard

• RAT-SQL

Live demoes

• Photon by the SalesForce group is a good live demo of text-to-SQL.