• Classification tasks Download a healthcare dataset e.g.: https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia

a)Activate Virtual Python Learning Environment b)Create Simulated PHRs for Patients: In the script 'data-partition.py' set the 'no_patients' variable to the number of patients you'd like to use in your experiment. -Set the location for the for the patient directories to be saved. -The script will attempt to equally divide the number of negative and positive classes to each patient.

c)Run python data-partition.py

Go to 'train.py' script. a)Import a TensorflowLite Model (e.g. a MobileNet version).

• Search for different model-types at https://tfhub.dev/ -Specify variable 'module_selection' with model variables,e.g. name,input size, etc) -Specify MODULE_HANDLE

b)Set FL-Variables in the 'train.py' script: 1.Batch_size = 2.num_epochs = 3.max_rounds = 4.no_patients =

a)Create a 'Global_model' subdirectory in a specified location to save the global_models output

• The Model will save in a sub-directory with the name of the FL-variables so that you can identify which model it was that was trained. b)Run 'python train.py' (advisable - advanced computing cluster)

Run export FL_CONFIG_FILE=/path/to/config/file In config file the following parameters are specified

• raw_data_root_dir : this is the place where the raw chest xray data set is stored prior to partition

• experiment_id: put any string without spaces over here to uniquely identify your experiment. could be anything as long as you can uniquely identify it examples names:

  1. num_patients and num_rounds, like "10_patients_5_rounds"
  2. data time of creation, like "17th_Oct_2021_17-56-09-UTC"
  3. unix timestamp of creation, like "1639531877"

• directories: the following are directory names.

  1. root_dir: the root directory where chestxray train, test and val folders are located
  2. data_root_dir: patient training data location name. for eg. if experiment_id = 1639531877, and root_dir: "/tmp/federated_learning_dataset/", then based on the below names data_root_directory is "/tmp/federated_learning_dataset/1639531877/data_root_dir/"
  3. test_dir: test data location name
  4. val_dir: validation data location name
  5. model_save_dir: model save directory name

Run data partitioner python data-partition.py that takes the config file from export FL_CONFIG_FILE=/path/to/config/file

Make sure mpi4py is installed. Refer to the following link

Use MPI to run mpirun -n <no_patients> python DistributedTrainer.py

• Setup private ipfs node: https://labs.eleks.com/2019/03/ipfs-network-data-replication.html
• Run ipfs daemon

• Export config file as above in ## Run Distributed Version

Run python ipfstrainer.py

• Install dependencies, geth, etc.
• Your IT cluster controller may have to install these for you

Run geth/go: srun -p interactive --pty /bin/bash module spider geth --> geth: geth/2022 module load geth/2022 on your computing cluster

• OR module load geth

Open two different windows for these commands. Window 1

• python regionNodeSetup.py 1 1
• python regionNodeSetup.py 1 0 Window 2
• python regionNodeConnector.py 1

Load mpi: module purge module load Anaconda3/2020.11 module load foss/2020a source activate $YOURLOCATION/mpienv

• run python submitContract.py
• run python smartContract_1.sol