Sawatabi is an application framework to develop and run stream-data-oriented Ising applications with quantum annealing.

The animation shows visualization of Outlier Detection Problem and its solutions obtained by a Sawatabi application. The Ising fomulation for the Outlier Detection Problem proposed in [1] is used. The left figure represents continuous input event data (stream-data) where blue points mean normal values and red points mean abnormal values (outliers). The right figure represents solutions for Outlier Detection for each window of the input stream-data. The boxes try to cover only normal points as much as possible.

[1] V. N. Smelyanskiy, E. G. Rieffel, S. I. Knysh, C. P. Williams, M. W. Johnson, M. C. Thom, W. G. Macready, and K. L. Pudenz, "A near-term quantum computing approach for hard computational problems in space exploration," arXiv:1204.2821 [quant-ph], 2012. Available: https://arxiv.org/abs/1204.2821

pip install sawatabi

This section only describes a sample application of NPP (Number Partition Problem), for other sample applications please see:
https://github.com/kotarot/sawatabi/tree/main/sample/algorithm

The following application reads numbers from a local file, run continuous annealing computations to solve NPP on local environment, and writes solutions to the stdout:

python sample/algorithm/npp_window.py --input="tests/algorithm/numbers_100.txt"

Please prepare your GCP service account credentials as ./gcp-key.json and open three terminals.

[1st terminal] The Pub/Sub publisher continuously publishes numbers to the specified Pub/Sub topic:

GOOGLE_APPLICATION_CREDENTIALS="./gcp-key.json" \
    python sample/pubsub/publish_pubsub.py \
        --project=<PROJECT> \
        --topic=<TOPIC> \
        --interval=1.0 \
        --random-number

where

• <PROJECT> is your GCP project name, and
• <TOPIC> is your Google Cloud Pub/Sub topic name to publish messages (numbers) to.

[2nd terminal] The Pub/Sub subscriber continuously subscribes solutions from the specified Pub/Sub subscription:

GOOGLE_APPLICATION_CREDENTIALS="./gcp-key.json" \
    python sample/pubsub/subscribe_pubsub.py \
        --project=<PROJECT> \
        --subscription=<SUBSCRIPTION>

where

• <PROJECT> is your GCP project name, and
• <SUBSCRIPTION> is your Google Cloud Pub/Sub subscription name to subscribe messages (solutions) from.

[3rd terminal] The following application reads numbers from the given Pub/Sub topic, run continuous annealing computations to solve NPP on Google Cloud Dataflow, and writes solutions to the given Pub/Sub topic:

GOOGLE_APPLICATION_CREDENTIALS="./gcp-key.json" \
    python sample/algorithm/npp_window.py \
        --project=<PROJECT> \
        --input-topic=<INPUT_TOPIC> \
        --output-topic=<OUTPUT_TOPIC> \
        --dataflow \
        --dataflow-bucket=<DATAFLOW_BUCKET>

where

• <PROJECT> is your GCP project name,
• <INPUT_TOPIC> is your Google Cloud Pub/Sub topic name of input,
• <OUTPUT_TOPIC> is your Google Cloud Pub/Sub topic name of output, and
• <DATAFLOW_BUCKET> is your GCS bucket name for Dataflow temporary files.

Please give credentials directly to the sawatabi.solver.DWaveSolver() constructor arguments, or set up a config using dwave-cloud-client:

$ dwave config create
Configuration file not found; the default location is: /path/to/your/location/dwave.conf
Configuration file path [/path/to/your/location/dwave.conf]:
Configuration file path does not exist. Create it? [y/N]: y
Profile (create new) [prod]: dev
API endpoint URL [skip]: xxxxxxxxxxxxxxxx
Authentication token [skip]: xxxxxxxxxxxxxxxx
Default client class [skip]:
Default solver [skip]: Advantage_system1.1
Configuration saved.

Please give credentials directly to the sawatabi.solver.OptiganSolver() constructor arguments, or set up a API Token in ~/.optigan.yml:

api:
    endpoint: http://optigan.example.com/method
    token: XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Please set up a development environment as follows:

python -m venv venv
source venv/bin/activate
pip install --upgrade pip
pip install --editable ".[dev]"

This work is supported by the MITOU Target program from Information-technology Promotion Agency, Japan (IPA).