Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Skip to content
/ TDCD Public

Code for Tiered Decentralized Coordinate Descent (TDCD)

2 stars 1 fork Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

rpi-nsl/TDCD

BranchesTags

Repository files navigation

Tiered Decentralized Coordinate-Descent (TDCD)

Dependencies

One can install our environment with Anaconda: conda env create -f flearn.yml

Dataset

ModelNet40 12-view PNG dataset can be downloaded from Google Drive. The code expects the dataset to be present in the current working directory in a folder named "view".

Prerequisite for Mimic3: Follow the instructions in mimic3-benchmark to download and pre-sanitize the datasets. After the extraction and sanitization is done, put the raw train, test and validation files in ~/data/mimi3lstm directory.

Run the batch scripts for running batch experiments

To run experiments using the batch scripts you can use the shell scripts we have provided.

For Cifar10 experiments, run:

./sbatch_cifar.sh

For Modelnet40 experiments, run:

./sbatch_modelnet.sh

For Mimic3 experiments, run:

./sbatch_mimic3.sh

Run the Cifar10 experiments individually using the python scripts

python cifar_journal.py [--seed [Seed to use]] [--hubs [Number of hubs/silos]]
                  [--clients [Number of clients in each silo]] 
                  [--gepochs [Number of global iterations]] 
                  [--Q [Number of local iterations per global iteration]]
                  [--batchsize [Overall number of samples selected in each iteration across all clients in each silo]] 
                  [--lr [Learning Rate of SGD]] 
                  [--momentum [Momentum in SGD (not used)]] 
                  [--evalafter [Evaluate after number of local iterations]]

Run the Modelnet40 experiments individually using the python scripts

python modelnet_journal.py [--seed [Seed to use]] [--hubs [Number of hubs/silos]]
                  [--clients [Number of clients in each silo]] 
                  [--gepochs [Number of global iterations]] 
                  [--Q [Number of local iterations per global iteration]]
                  [--batchsize [Overall number of samples selected in each iteration across all clients in each silo]] 
                  [--lr [Learning Rate of SGD]] 
                  [--momentum [Momentum in SGD (not used)]] 
                  [--evalafter [Evaluate after number of local iterations]]
                  [--modelnet_type [Determine the type of Modelnet used. Possible values ModelNet10 or ModelNet40.]]

Run the Mimic3 experiments individually using the python scripts

python mimic3_journal.py [--seed [Seed to use]] [--hubs [Number of hubs/silos]]
                  [--clients [Number of clients in each silo]] 
                  [--gepochs [Number of global iterations]] 
                  [--Q [Number of local iterations per global iteration]]
                  [--batchsize [Overall number of samples selected in each iteration across all clients in each silo]] 
                  [--lr [Learning Rate of SGD]] 
                  [--momentum [Momentum in SGD (not used)]] 
                  [--evalafter [Evaluate after number of local iterations]]

Sample plot generating scripts are provided in `~/plot_utils/

About

Code for Tiered Decentralized Coordinate Descent (TDCD)

Resources

Readme
Activity
Custom properties

Stars

2 stars

Watchers

2 watching

Forks

1 fork
Report repository

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages