Optimizaion of MLL by fit_gpytorch_mll with Adam algorithm

[MJeong19]

Currently, I am trying to use Adam algorithm to optimize marginal log likelihood.
However, an error that the object is not callable occured.
This error still occured after changing params=mll.parameters into model.parameters.
Here is the code and error occured:

iteration = 10
trainx = t #tensor
trainy = obj2 #tensor
mgl= MultitaskGaussianLikelihood(num_tasks=6)
covar_module = MaternKernel(nu=2.5, ard_num_dims=6)

model = SingleTaskGP(train_X = trainx, train_Y = trainy, likelihood= mgl, covar_module = covar_module, outcome_transform=Standardize(m=6))
mll = ExactMarginalLogLikelihood(mgl, model)
fit_gpytorch_mll(mll, optimizer=torch.optim.Adam(params=mll.parameters(), lr=0.001))

Why this error occurs, and what should I do to solve this error?

[Balandat]

optimizer is supposed to be a botorch optimizer callable not a torch Optimizer object, so fit_gpytorch_mll_scipy or fit_gpytorch_mll_torch as defined in https://github.com/pytorch/botorch/blob/4632c36bf6f74387b9a7064c2ce491a5a3d22f95/botorch/optim/fit.py (or if you want to implement a custom one you'll need to define a function complying with their API).

fit_gpytorch_mll_torch will use Adam by default. If you want to choose custom options you can pass a custom make_torch_optimizer() factory function as the "optimizer_kwargs: {"optimizer": make_torch_optimizer}" to fit_gpytorch_mll that will be used here:

botorch/botorch/optim/fit.py

Lines 141 to 142 in 4632c36

	optimizer: A `torch.optim.Optimizer` instance or a factory that takes
	a list of parameters and returns an `Optimizer` instance.

Do you have a particular reason to use Adam here?