distFedPAQ-simulation / README.md
README.md
Raw

distFedPAQ-simulation

About

You can find this repo at https://github.com/aheritianad/distFedPAQ-simulation.

A copy is available here https://gitfront.io/r/aheritianad/Jf8eQDMabdcc/distFedPAQ-simulation/.

Installation

  1. Get a copy of this repo by

  2. Enter into the directory
    cd distFedPAQ-simulation

  3. [OPTIONAL but RECOMMENDED] Use a virtual environment

    • Make a virtual environment .venv with

      python3 -m venv .venv

    • Activate the virtual environment with

      source .venv/bin/activate

      You can deactivate .venv anytime with the deactivate command.

  4. Install all dependencies

    pip3 install -r ./distFedPAQ/requirements.txt

Usage

Commands

  1. See help for the arguments

    python3 simulator.py --help or python3 simulator.py -h

  2. Example of a simulation command

    python -i simulator.py -n 5 -loc 10 -ext 100 -n-ext-ave 2 -bs 12 -lr 1e-1 -mom 0 -tr seq -P-path proba/ring_5.csv -rs 5

I got the following result by running the above command on diabetes of scikit-learn:

Code Plots Note
plot(stop=-1) sample_simulation All in one
plot(stop=-1,test=False) train On train dataset
plot(stop=-1,train=False) test On test dataset
plot(1, single=False) node 1 $Node_{1}$
plot(start=2, stop=-1, single=False, ave=False, test=False) node 2 to -1 from $Node_{2}$ to the last node (i.e. $Node_{5}$) on the train dataset
plot_graph(show_contact=False) graph without contacts Graph whit only probabilities.
plot_graph() graph with contacts Graph with probabilities and number of contacts between nodes during training.

Arguments

Required

Short Long Description
-n --nodes Number of nodes.

Optional

Short Long Description Default value
-loc --local-update Number of iterations per node for each local update. 10
-n-ext-ave --nodes-external-averaging Number nodes will update externally by averaging. 2
-ext --external-update Number of time that some nodes (see -n-ext-ave) will update externally by averaging. 100
-bs --batch-size Size of a batch from local data of a node in each iteration of its local update. 1
-lr --learning-rate Learning rate for nodes' local update. 1e-3
-mom --momentum Momentum for a local update. Value in [0,1). 0
-bias --with-bias Flag for using bias for the model [y/n]. y
-tr --trainer Trainer for the nodes [[seq/sequential]/[par/parallel]]. seq
-rs --repeat-single Number of time that the single node will do loops of local updates before evaluation. 1
-P-path --path-to-probability-P Path to the probability file (csv type) which encode the graph. If not set, probability will be uniform on all nodes. P will be a uniform.

Notes

  1. I recommend to use -tr seq when -loc is small (<=3000) and -tr par otherwise.
  2. In order to get more flexibility, it is recommended to run in interactive mode (i.e. python3 -i ...).

Author

Heritiana Daniel Andriasolofo

Powered by GitFront