How to use Exogenous Variables with AutoNHITS or AutoTFT?

[blackhole-canada]

Hello, I am trying to use some future, historic and static exogenous variables with NHITS and TFT models while doing hyperparameter optimization. However, I get an error about unexpected arguements while using exogenous variables with automodels while it works fine for normal NHITS or TFT models. Is there a way to use the exogenous variables for the auto models? Thank you in advance.

[elephaint]

Hi, you can do:

config = dict(input_size=12, hist_exog_list=[], stat_exog_list=[], futr_exog_list=[])
model = AutoNHITS(h=12, config=config)

[supriamir]

Hi @elephaint, i want to ask regarding to fine tuning and exogenous variables,

nhits_config = {
"learning_rate": tune.choice([1e-3]),
"max_steps": tune.choice([1000]),
"input_size": tune.choice([2 * horizon]),
"batch_size": tune.choice([7]),
"windows_batch_size": tune.choice([256]),
"n_pool_kernel_size": tune.choice([[2, 2, 2], [16, 8, 1]]),
"n_freq_downsample": tune.choice([[168, 24, 1], [24, 12, 1], [1, 1, 1]]),
"activation": tune.choice(['ReLU']),
"n_blocks": tune.choice([[1, 1, 1]]),
"mlp_units": tune.choice([[[512, 512], [512, 512], [512, 512]]]),
"interpolation_mode": tune.choice(['linear']),
"val_check_steps": tune.choice([100]),
"random_seed": tune.randint(1, 10),
}

models = [
AutoTFT(
h=horizon,
loss=MAE(),
config=config,
num_samples=10,
search_alg=HyperOptSearch(),
backend='ray',
),
]

How can i use fine tuning and exogenous variable at the same time? Because in documentation, parameter config is used for hyperparameter optimization.

[blackhole-canada]

Thank you so much, it worked fine :)