Update document and the set up script

CI-install.sh

@@ -1,24 +1,17 @@
 #!/bin/bash
 set -ex
 
-# install pyflakes to do code error checking
-echo "pip3 install pyflakes --cache-dir $HOME/.pip-cache"
-pip3 install pyflakes --cache-dir $HOME/.pip-cache
+source "$HOME/miniconda3/etc/profile.d/conda.sh"
 
-# Conda Installation
-wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
-bash miniconda.sh -b -p $HOME/miniconda
-source "$HOME/miniconda/etc/profile.d/conda.sh"
-hash -r
-conda config --set always_yes yes --set changeps1 no
+conda config --set always_yes yes #--set changeps1 no
 conda update -q conda
 conda info -a
 
-echo "conda create -n fedml python=3.7.4"
-conda create -n fedml python=3.7.4
+echo "conda create -n feddrift python=3.7.4"
+conda create -n feddrift python=3.7.4
 
-echo "conda activate fedml"
-conda activate fedml
+echo "conda activate feddrift"
+conda activate feddrift
 
 # Install PyTorch (please visit pytorch.org to check your version according to your physical machines
 conda install pytorch torchvision cudatoolkit=10.2 -c pytorch
@@ -42,50 +35,3 @@ cd ./../../../
 
 # Install wilds package (still have to download the dataset separately)
 pip install wilds
-
-# install the dataset
-# 1. MNIST
-cd ./data/MNIST
-sh download_and_unzip.sh
-cd ../../
-
-# 2. FederatedEMNIST
-cd ./data/FederatedEMNIST
-sh download_federatedEMNIST.sh
-cd ../../
-
-# 3. shakespeare
-cd ./data/shakespeare
-sh download_shakespeare.sh
-cd ../../
-
-
-# 4. fed_shakespeare
-cd ./data/fed_shakespeare
-sh download_shakespeare.sh
-cd ../../
-
-# 5. fed_cifar100
-cd ./data/fed_cifar100
-sh download_fedcifar100.sh
-cd ../../
-
-# 6. stackoverflow
-cd ./data/stackoverflow
-sh download_stackoverflow.sh
-cd ../../
-
-# 7. CIFAR10
-cd ./data/cifar10
-sh download_cifar10.sh
-cd ../../
-
-# 8. CIFAR100
-cd ./data/cifar100
-sh download_cifar100.sh
-cd ../../
-
-# 9. CINIC10
-cd ./data/cinic10
-sh download_cinic10.sh > cinic10_downloading_log.txt
-cd ../../

DATA-install.sh

@@ -0,0 +1,47 @@
+
+# install the dataset
+# 1. MNIST
+cd ./data/MNIST
+sh download_and_unzip.sh
+cd ../../
+
+# 2. FederatedEMNIST
+cd ./data/FederatedEMNIST
+sh download_federatedEMNIST.sh
+cd ../../
+
+# 3. shakespeare
+cd ./data/shakespeare
+sh download_shakespeare.sh
+cd ../../
+
+
+# 4. fed_shakespeare
+cd ./data/fed_shakespeare
+sh download_shakespeare.sh
+cd ../../
+
+# 5. fed_cifar100
+cd ./data/fed_cifar100
+sh download_fedcifar100.sh
+cd ../../
+
+# 6. stackoverflow
+cd ./data/stackoverflow
+sh download_stackoverflow.sh
+cd ../../
+
+# 7. CIFAR10
+cd ./data/cifar10
+sh download_cifar10.sh
+cd ../../
+
+# 8. CIFAR100
+cd ./data/cifar100
+sh download_cifar100.sh
+cd ../../
+
+# 9. CINIC10
+cd ./data/cinic10
+sh download_cinic10.sh > cinic10_downloading_log.txt
+cd ../../

README.md

@@ -11,6 +11,56 @@ Empirical evaluation shows that our solutions achieve significantly higher accur
 
 This repository is built on top of a federated learning research platform, [FedML](https://github.com/FedML-AI/FedML). 
 
+# Setup Environment
+
+Our installation script is based on [Miniconda](https://docs.conda.io/projects/miniconda/en/latest/). Please modify the script according to your package manager.
+
+```
+$ ./CI-install.sh
+```
+
+Our expeirments are tested with Python 3.7.4 and PyTorch 10.2. You can also install the GPU-enabled PyTorch by modifying the script above.
+
+The above script will create a conda environment named `feddrift` and install all the required packages. Please activate the environment before running any experiments.
+
+```
+$ conda activate feddrift
+```
+
+# Running Experiments
+
+We use [Weights & Biases](https://wandb.ai/site) to log experiments. Please create an account and log in before running an experiment.
+
+```
+$ wandb login
+```
+
+To run an experiment, please use the following command:
+
+```
+$ cd fedml_experiments/distributed/fedavg_cont_ens
+$ ./run_fedavg_distributed_pytorch.sh ${CLIENT_NUM} ${WORKER_NUM} ${SERVER_NUM} ${GPU_PER_SERVER} ${MODEL} ${DATA_DIST} ${ROUND} ${EPOCH} ${BATCH_SIZE} ${LR} ${DATASET} ${DATA_DIR} ${SAMPLE_NUM} ${NOISE_PROB} ${ONLY_TEST_ONE_CLIENT} ${TOTAL_ITER} ${CONCEPT_NUM} ${RESET_MODEL} ${DRIFT_TOGETHER} ${DRIFT_ALGO} ${DRIFT_ALGO_ARG} ${TIME_STRETH} ${DUMMY_ARG} ${CHANGE_POINT}
+```
+
+For example, running FedDrift for SEA-4 dataset and change point **A**:
+
+```
+$ ./run_fedavg_distributed_pytorch.sh 10 10 1 4 fnn homo 200 5 500 0.01 sea "./../../../data/" 100 0 0 10 4 0 0 softcluster H_A_C_1_10_0 1 0 A
+```
+
+The major algorithms we implemented are listed here (in the form of `${DRIFT_ALGO}` + optionally `${DRIFT_ALGO_ARG}`):
+
+* `win-1`: Window method (one time step)
+* `win-2`: Window method (two time step)
+* `all`: Oblivous method (using all training data)
+* `dsurf`: [DriftSurf](https://proceedings.mlr.press/v139/tahmasbi21a/tahmasbi21a.pdf)
+* `aue`: [Accuracy Updated Ensemble](https://link.springer.com/chapter/10.1007/978-3-642-21222-2_19)
+* `kue`: [Kappa Updated Ensemble](https://link.springer.com/article/10.1007/s10994-019-05840-z)
+* `softcluster` + `cfl_0.1_win-1`: Clustered federated learning
+* `softclusterwin-1` + `hard-r`: [IFCA](https://arxiv.org/abs/2006.04088)
+* `ada` + `win-1_iter` [Adaptive-FedAvg](https://ieeexplore.ieee.org/document/9533710)
+
+
 # Reference Papers
 
 If you use our code in your work, we would appreciate a reference to the following papers

fedml_mobile/server/executor/requirements.txt

@@ -14,4 +14,4 @@ paho-mqtt==1.5.0
 Celery==5.2.2
 pandas==1.1.3
 setproctitle
-sklearn
+