Skip to content

Latest commit

 

History

History
242 lines (163 loc) · 9.74 KB

collaborative_learning.rst

File metadata and controls

242 lines (163 loc) · 9.74 KB
Raw

Collaborative Learning

Background

This demo shows a user_manual_scenarios_collaborative_learning and the nodes involved: user_manual_nodes_model_receiver and user_manual_nodes_model_sender. With these 2 nodes implemented, the user can deploy as many nodes of each kind as desired and check the behavior of a simulated network running. They are implemented in Python to prove the communication between the 2 implementations.

The purpose of the demo is to show how a Sender and a Receiver node can communicate. The Receiver node awaits model statistics from the Sender. Since the Sender doesn't have a real AML Engine, it sends the model statistics as a string. Upon receiving the statistics, the Receiver sends a model request, also as a string since it doesn't have an AML Engine. Then, the Sender converts the received model request to uppercase and sends it back as a model reply.

Prerequisites

Before running this demo, ensure that AML-IP is correctly installed using one of the following installation methods:

  • installation_manual_linux
  • installation_manual_windows
  • docker

Building the demo

If the demo package is not compiled, please refer to developer_manual_installation_sources_linux_colcon_demos or run the command below.

colcon build --packages-up-to amlip_demo_nodes

Once AML-IP packages are installed and built, import the libraries using the following command.

source install/setup.bash

Explaining the demo

In this section, we will delve into the details of the demo and how it works.

Model Manager Receiver Node

This is the Python code for the user_manual_nodes_model_receiver application. It does not use real AML Models, but strings. It is implemented in using amlip_py API.

This code can be found here.

The next block includes the Python header files that allow the use of the AML-IP Python API.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_receiver_custom.py

Let's continue explaining the global variables.

DOMAIN_ID variable isolates the execution within a specific domain. Nodes with the same domain ID can communicate with each other.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_receiver_custom.py

waiter is a WaitHandler that waits on a boolean value. Whenever this value is True, threads awake. Whenever it is False, threads wait.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_receiver_custom.py

The CustomModelListener class listens to user_manual_scenarios_collaborative_learning_statistics and user_manual_scenarios_collaborative_learning_model_reply messages received from a user_manual_nodes_model_sender. This class is supposed to be implemented by the user in order to process the messages received from other nodes in the network.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_receiver_custom.py

The main function orchestrates the execution of the Model Manager Receiver node. It creates an instance of the ModelManagerReceiverNode and starts its execution with the specified listener.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_receiver_custom.py

After starting the node, it waits for statistics to arrive from the user_manual_nodes_model_sender.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_receiver_custom.py

Then, it requests a model from the user_manual_nodes_model_sender using the received server ID.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_receiver_custom.py

Finally, the node stops.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_receiver_custom.py

Model Manager Sender Node

This is the Python code for the user_manual_nodes_model_sender application. It does not use real AML Models, but strings. It does not have a real AML Engine but instead the calculation is an upper-case conversion of the string received. It is implemented in using amlip_py API.

This code can be found here.

The following block includes the Python header files necessary for using the AML-IP Python API.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

Let's continue explaining the global variables.

DOMAIN_ID isolates the execution within a specific domain. Nodes with the same domain ID can communicate with each other.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

waiter is a WaitHandler that waits on a boolean value. Whenever this value is True, threads awake. Whenever it is False, threads wait.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

The CustomModelReplier class listens to user_manual_scenarios_collaborative_learning_model_request request messages received from a user_manual_nodes_model_receiver. This class is supposed to be implemented by the user in order to process the messages.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

The main function orchestrates the execution of the Model Manager Sender node. It creates an instance of ModelManagerSenderNode.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

After starting the node, it publishes statistics using the publish_statistics() function, which fills a user_manual_scenarios_collaborative_learning_statistics and publishes it.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

Then we start the node execution, passing the previously defined CustomModelReplier() class, which is responsible for managing the request received.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

Waits for the response model to be sent to the user_manual_nodes_model_receiver.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

Finally, it stops and closes the node.

/../amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo/model_sender_custom.py

Running the demo

This demo runs the implemented nodes in amlip_demo_nodes/amlip_collaborative_learning_demo.

Run Model Manager Receiver Node

Run the following command:

# Source colcon installation
source install/setup.bash

# To execute Model Manager Receiver Node
cd ~/AML-IP-ws/src/AML-IP/amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo
python3 model_receiver_custom.py

The expected output is the following:

Starting Manual Test Model Manager Receiver Node Py execution. Creating Node...
Node created: ModelManagerReceiver.0f.19.23.2d. Already processing models.
Model reply received from server
solution: MOBILENET V1
Finishing Manual Test Model Manager Receiver Node Py execution.

Run Model Manager Sender Node

Run the following command to answer before closing:

# Source colcon installation
source install/setup.bash

# To execute Model Manager Sender Node
cd ~/AML-IP-ws/src/AML-IP/amlip_demo_nodes/amlip_collaborative_learning_demo/amlip_collaborative_learning_demo
python3 model_sender_custom.py

This execution expects an output similar to the one shown below:

Starting Manual Test Model Manager Sender Node Py execution. Creating Node...
Node created: ModelManagerSender.0a.14.1e.28. Already processing models.
Model request received from client
model: MobileNet V1
solution: MOBILENET V1
Finishing Manual Test Model Manager Sender Node Py execution.