standard experiment

.gitignore

@@ -9,5 +9,8 @@
 **.egg-info
 2021**
 2022**
+2024**
+masters-mathis/*
+eval/CIFAR/*
 **/massif.out*
 *swp

eval/35_workers.edges

+35
+0 17
+0 34
+0 30
+0 15
+1 20
+1 12
+1 5
+1 6
+2 10
+2 34
+2 14
+2 23
+3 8
+3 33
+3 29
+3 23
+4 16
+4 11
+4 28
+4 29
+5 1
+5 23
+5 14
+5 31
+6 1
+6 28
+6 22
+6 9
+7 18
+7 28
+7 21
+7 15
+8 26
+8 3
+8 22
+8 15
+9 10
+9 19
+9 29
+9 6
+10 16
+10 9
+10 18
+10 2
+11 24
+11 27
+11 4
+11 22
+12 1
+12 26
+12 17
+12 31
+13 24
+13 17
+13 34
+13 21
+14 32
+14 2
+14 5
+14 21
+15 8
+15 0
+15 22
+15 7
+16 25
+16 34
+16 10
+16 4
+17 0
+17 18
+17 12
+17 13
+18 17
+18 10
+18 27
+18 7
+19 9
+19 20
+19 30
+19 25
+20 32
+20 25
+20 19
+20 1
+21 28
+21 13
+21 14
+21 7
+22 8
+22 11
+22 6
+22 15
+23 32
+23 2
+23 3
+23 5
+24 11
+24 27
+24 13
+24 31
+25 16
+25 33
+25 19
+25 20
+26 8
+26 33
+26 12
+26 30
+27 24
+27 18
+27 11
+27 31
+28 4
+28 21
+28 6
+28 7
+29 9
+29 3
+29 4
+29 33
+30 0
+30 32
+30 26
+30 19
+31 24
+31 27
+31 12
+31 5
+32 20
+32 23
+32 30
+32 14
+33 25
+33 26
+33 3
+33 29
+34 16
+34 0
+34 2
+34 13

src/decentralizepy/node/DPSGDNode.py

@@ -3,6 +3,7 @@ import json
 import logging
 import math
 import os
+import pickle
 from collections import deque
 
 import torch
@@ -68,6 +69,9 @@ class DPSGDNode(Node):
         global_epoch = 1
         change = 1
 
+        with open("flweights.pkl", "rb") as f:
+            self.model.load_state_dict(pickle.load(f))
+
         for iteration in range(self.iterations):
             logging.info("Starting training iteration: %d", iteration)
             rounds_to_train_evaluate -= 1

src/decentralizepy/node/FederatedParameterServer.py

@@ -3,6 +3,7 @@ import json
 import logging
 import math
 import os
+import pickle
 import random
 from collections import deque
 
@@ -277,6 +278,11 @@ class FederatedParameterServer(Node):
         to_send = dict()
 
         for iteration in range(self.iterations):
+
+            if iteration == 1000:
+                with open("flweights.pkl", "wb") as f:
+                    pickle.dump(self.model.state_dict(), f)
+
             self.iteration = iteration
             # reset deques after each iteration
             self.peer_deques = dict()

tutorial/config_celeba_sharing.ini

@@ -2,8 +2,8 @@
 dataset_package = decentralizepy.datasets.CIFAR10
 dataset_class = CIFAR10
 model_class = LeNet
-train_dir = /mnt/nfs/shared/CIFAR
-test_dir = /mnt/nfs/shared/CIFAR
+train_dir = ./CIFAR
+test_dir = ./CIFAR
 ; python list of fractions below
 sizes = 
 random_seed = 90
@@ -28,7 +28,7 @@ loss_class = CrossEntropyLoss
 [COMMUNICATION]
 comm_package = decentralizepy.communication.TCP
 comm_class = TCP
-addresses_filepath = /mnt/nfs/risharma/Gitlab/tutorial/ip.json
+addresses_filepath = ../tutorial/ip.json
 
 [SHARING]
 sharing_package = decentralizepy.sharing.Sharing

tutorial/run_decentralized.sh

 #!/bin/bash
 
-decpy_path=/mnt/nfs/risharma/Gitlab/decentralizepy/eval
+decpy_path=../eval
 cd $decpy_path
 
-env_python=~/miniconda3/envs/decpy/bin/python3
-graph=/mnt/nfs/risharma/Gitlab/tutorial/96_regular.edges
-original_config=/mnt/nfs/risharma/Gitlab/tutorial/config_celeba_sharing.ini
-config_file=~/tmp/config.ini
-procs_per_machine=16
+env_python=python3
+graph=./35_workers.edges
+original_config=../tutorial/config_celeba_sharing.ini
+config_file=config.ini
+procs_per_machine=35
 machines=1
-iterations=80
-test_after=20
-eval_file=testingPeerSampler.py
+iterations=2000
+test_after=50
+eval_file=testing.py
 log_level=INFO
 
-m=`cat $(grep addresses_filepath $original_config | awk '{print $3}') | grep $(/sbin/ifconfig ens785 | grep 'inet ' | awk '{print $2}') | cut -d'"' -f2`
+m=0
 echo M is $m
 log_dir=$(date '+%Y-%m-%dT%H:%M')/machine$m
 mkdir -p $log_dir

tutorial/run_federated.sh

 #!/bin/bash
 
-decpy_path=/mnt/nfs/risharma/Gitlab/decentralizepy/eval
+decpy_path=../eval
 cd $decpy_path
 
-env_python=~/miniconda3/envs/decpy/bin/python3
-graph=/mnt/nfs/risharma/Gitlab/tutorial/96_regular.edges
-original_config=/mnt/nfs/risharma/Gitlab/tutorial/config_celeba_sharing.ini
-config_file=~/tmp/config.ini
-procs_per_machine=16
+env_python=python3
+graph=./35_workers.edges
+original_config=../tutorial/config_celeba_sharing.ini
+config_file=config.ini
+procs_per_machine=35
 machines=1
-iterations=80
-test_after=20
+iterations=2000
+test_after=50
 eval_file=testingFederated.py
 log_level=INFO
 server_rank=-1
 server_machine=0
-working_rate=0.5
+working_rate=1.0
 
-m=`cat $(grep addresses_filepath $original_config | awk '{print $3}') | grep $(/sbin/ifconfig ens785 | grep 'inet ' | awk '{print $2}') | cut -d'"' -f2`
+m=0
 echo M is $m
 log_dir=$(date '+%Y-%m-%dT%H:%M')/machine$m
 mkdir -p $log_dir
 
 cp $original_config $config_file
 # echo "alpha = 0.10" >> $config_file
-$env_python $eval_file -ro 0 -tea $test_after -ld $log_dir -mid $m -ps $procs_per_machine -ms $machines -is $iterations -gf $graph -ta $test_after -cf $config_file -ll $log_level -ctr 0 -cte 0 -wsd $log_dir -sm $server_machine -sr $server_rank -wr $working_rate
\ No newline at end of file
+$env_python $eval_file -ro 0 -tea $test_after -ld $log_dir -mid $m -ps $procs_per_machine -ms $machines -is $iterations -gf $graph -ta $test_after -cf $config_file -ll $log_level -wsd $log_dir -sm $server_machine -sr $server_rank -wr $working_rate
\ No newline at end of file