Quickstart

Starting using the STADLE platform is as easy as just following the 3 steps below.

STEP 1: Set up STADLE Server

Just go to the stadle.ai and sign up for a free account. Then, login to your account, create a new project, and initiate an aggregator. Just wait for a few seconds to get the aggregator running.

Please also follow the step-by-step instructions on the Start Yout Project page in User Guide.

Note

With your free account, you will be able to create one project and initiate one aggregator. The number of agents that can be connected to the aggregator will be also limited.

STEP 2: Install STADLE Client

The STADLE client can be installed using the following commands.

First upgrade the pip.

pip install --upgrade pip

Then, install the stadle-client.

pip install stadle-client

Note

The environment needs to be Python 3.8.0 or newer.

STEP 3: Run Local STADLE Example Codes

The next phase is to utilize STADLE libraries by importing them in your local ML processes. You can quickly run and test some of the sample applications from STADLE examples in which some of the key STADLE client libraries are already connected. To do so, just download the local example codes from the repo.

git clone https://github.com/tie-set/stadle_examples.git

After downloading the sample codes, just follow the instruction of the README on how to run those applications with the STADLE client-side APIs.

For example, to run the minimal example using PyTorch, just go to the “stadle_examples/minimal_examples/pytorch” folder. Then, modify the config files for both admin and ML agents. In particular, “aggr_ip” should be the “IP Address” and “reg_port” should be the “Port to Connect”, both shown on the STADLE dashboard.

Then, run the admin agent with the following command (In this case, the script is named minimal_admin_agent.py). The admin agent uploads the base model that defines the architecture of the ML models that will be aggregated.

python minimal_admin_agent.py

Then, go to the STADLE dashboard and update the page after a few seconds. You can check the name of the uploaded base model on the dashboard. You can run multiple agents with different agent names. In this case, the name of the local ML process script is “minimal_fl_agent.py”. For example, you can run the ML agents like

python minimal_fl_agent.py --agent_name AGEMT_NAME_01
python minimal_fl_agent.py --agent_name AGEMT_NAME_02

On the STADLE dashboard, you will see the number of connected ML agents and downloadable recent global and local models as well as the best performing model.

You have successfully completed all the steps to conduct a STADLE project properly. The aggregation process for each round can be checked and managed on Aggregation Management page. Also, the configuration information and system status of aggregators and agents can be checked on Config Info & Settings page.

You will also be able to download the recent global, local, best performance models as well as keep track of the performance of ML models on the STADLE dashboard.