My experiments are runs of different models over different datasets. Different models for example have their own sets of hyperparameters, same for datasets.
For example for CNN model, I want to save kernel size and for Transformer model its residual dimension. For datasets I want to save their preprocessing parameters, which could be multiple.
That is,
config1 = {
"model": {
"name": "CNN",
"kernel_size": 3
}
"dataset": {
"name": "MNIST",
"preprocessing": None
}
}
config2 = {
"model": {
"name": "Transformer",
"d_model": 512
}
"dataset": {
"name": "MNIST",
"preprocessing": [
"noise": {
"mean": 0.0,
"std": 1.0
},
"resize": 14
]
}
}
Would I able to do this with W&B? Examples in documentation are all plain dictionaries.