Job - input.json

This page contains an in-depth description on the input.json file

With the help of the previous section, you can view available sites and data description across sites using which you can define a job in the following manner.

Parameter definitions

1. name (str) - Defines the name of the job you want to submit

2. task (str) - It can only take two values - train or validate.

3. sites (str) - It contains the site names seperated by comma (,) with no spaces. For eg. "site-1,site-2"

4. rounds (str) - It defines the number of federated rounds you want the model to take. During each round of federated learning, the data sites update the model using their local data and send the updated model back to the server. You can checkout the paper on FedAvg algorithm to understand deeper into what one round represents.

5. nnClass (str) - It defines the name of the neural network class as defined in nn.py file.

6. Site-specific JSON (dict) - This defines the deep learning parameters across each data site. Each site will have a seperate dict defining the parameters for that site specifically. For getting available sites and data description of each site, please refer to the previous section - Preview Site Description. Inside this site specific dict, the following parameters are defined:

   1. aggregation_epochs (int) - It defines the the number of epochs to run for each update before sending to server.

   2. lr (float) - Learning Rate for the optimizer defined in nnMetrics.py.

   3. batch_size (int) - It defines the local batch size

   4. data_size (int) - It defines the the size of the data subset you want to choose from the overall data at the site.

   5. train_test_split (float) - It defines the train_test_split ratio of the chosen data subset. Its value can be between 0 and 1.

NOTE 1: If task:"train", the site parameters will include all the mentioned parameters above. However, if task:"validate", the site parameters should include batch_size and data_size. Rest of the parameters do not apply for only validation.

NOTE 2: In the free trial version with public sites, please keep the number of data points less than 50 at each site and aggregation epochs less than 5, otherwise the application will fail to run. In the paid version with private sites, there are no threshold restrictions.

Sample train file

{
    "name": "TrainCNN_V1",
    "task": "train",
    "sites": "site-1,site-2",
    "rounds": "2",
    "nnClass": "MobileNetCNN",
    "site-1": {
        "aggregation_epochs": 2,
        "lr": 0.01,
        "batch_size": 8,
        "data_size": 15,
        "train_test_split": 0.3,
        "balanced_class_train": "true"
    },
    "site-2": {
        "aggregation_epochs": 2,
        "lr": 0.01,
        "batch_size": 8,
        "data_size": 15,
        "train_test_split": 0.3,
        "balanced_class_train": "true"
    }
}

Sample validate file

{
    "name": "ValidateCNN_V1",
    "task": "validate",
    "sites": "site-1,site-2",
    "rounds": "2",
    "nnClass": "MobileNetCNN",
    "site-1": {
        "batch_size": 8,
        "data_size": 15
    },
    "site-2": {
        "batch_size": 8,
        "data_size": 25
    }
}